00 The Future of Drug Innovation

The future of drug innovation
Supplement to Nature Publishing Group Journals October 2011

reprint collection
S P ONS OR E D CONT E NT
www.nature.com/reprintcollections/lilly/drug-innovation Sponsored by
For the past 135 years, science has been at the foundation of all we do
here at Eli Lilly and Company. We adapt and evolve in order to research
and develop breakthrough medicines that deliver improved outcomes
for individual patients. From collaborating to produce the worlds frst
insulin, to the discovery and development of medicines that have helped
revolutionize the treatment of depression and schizophrenia, Lilly has
spent decades delivering innovative medicines that address unmet
medical needs in neuroscience, diabetes, cancer, osteoporosis and more.
We are currently standing at the threshold of potential new breakthroughs
in science and we plan to deliver to the patients who are waiting.
INSIDE Lef Cover
www.nature.com/reprintcollections/lilly/drug-innovation NATURE REPRINT COLLECTION | THE FUTURE OF DRUG INNOVATION | OCTOBER 2011 | 1
Nature Reprint Collection
publisher: MELANIE BRAZIL
sponsorship: DAVID BAGSHAW, YVETTE SMITH,
GERARD PRESTON
web production: ANITHA SONIA, MANPREET MANKOO
manufacturing production: SUSAN GRAY,
STEPHENRUSSELL
cover artist: SUSIE LANNI
sponsor: ELI LILLY AND COMPANY
CITING THE COLLECTION
All papers have been previously published in Nature
Publishing Group journals. Please use the original citation,
which can be found on the table of contents below.
For the past 135 years, science has been at the foundation of all we do
here at Eli Lilly and Company. We adapt and evolve in order to research
and develop breakthrough medicines that deliver improved outcomes
for individual patients. From collaborating to produce the worlds frst
insulin, to the discovery and development of medicines that have helped
revolutionize the treatment of depression and schizophrenia, Lilly has
spent decades delivering innovative medicines that address unmet
medical needs in neuroscience, diabetes, cancer, osteoporosis and more.
We are currently standing at the threshold of potential new breakthroughs
in science and we plan to deliver to the patients who are waiting.
INSIDE Lef Cover
Supplement to Nature Publishing Group Journals October 2011
reprint collection
S P ONS OR E D CONT E NT
www.nature.com/reprintcollections/lilly/drug-innovation Sponsored by
Sponsors Foreword
T
he biopharmaceutical industry faces intense pressures including scientifc, regulatory
and payer-related concerns that afect R&D productivity. Just as signifcant are the
opportunities to serve unmet patient needs. The worlds population is also aging and,
as a result, the number of patients with Alzheimers disease, diabetes and cancer, and other
chronic diseases associated with aging is increasing. There is an urgent need for medicines
that cure, prevent, or slow the progression of disease.
To provide innovative treatments that address diseases caused by interplay between
genetics, lifestyle, and environmental factors, Lilly, and the rest of the biopharmaceutical
industry, will need to continue to discover and develop new medicines often collaborating
with academia, biotechnology companies, and regulatory agencies to do so. In fact, the
current diabetes mellitus type 2 global epidemic is one key example of a complex disease
involving factors such as lack of exercise and inappropriate nutrition, coupled with genetic
susceptibility.
The articles included in this supplement demonstrate some of the problems facing the
biopharmaceutical industry today, but solutions are also included which focus on answers
to critical questions like: How can we intervene to treat diseases earlier or prevent disease
progression? How can genome mining be done more efectively in order to fnd more human
validated disease targets and increase our understanding of multiple cell signaling pathways?
Other aspects of the solution the industry is pursuing include the development of
multi-specifc therapeutics to boost efcacy by combining more than one unique mechanism
of action into one molecule. Industry is increasingly tailoring medicines to suit specifc
patient populations, including the use of companion diagnostics. Therefore, at Lilly, we are
developing a deeper understanding of genetic diferences between ethnic patient groups, a
prerequisite for global registrations. We are also increasingly better at understanding the real
life patient experience with our medicines, including the importance of patient convenience
for delivery and administration.
Through these eforts to improve success rates and other strategies, Lilly will improve R&D
productivity and shift away from the current inefcient paradigm of multiple shots on goal
to try to get medicines to market. Rather, we like the idea of timely shots in goal, which
means delivering a timely fow of innovative medicines with a greater likelihood of reaching
patients with a diferentiated value proposition for clinically meaningful outcomes including
data with relevant comparators. This evolution in our R&D strategy will beneft patients,
prescribers, payers, and other stakeholders and fuel our industry in the years ahead.
The future advances of treatments for human disease will continue to be determined, to
a large extent, by the innovation generated by the biopharmaceutical industry and in close
collaboration with others. Vast knowledge and experience with disease mechanisms and
therapeutic molecule research and development are held within biopharma companies.
Successful treatment advances will also depend on a healthcare innovation ecosystem that
encourages continuous enhancements of scientifc, technical and medical capabilities. This
is supported by advanced analytic information technology, as well as productive external
partnerships to access unique capabilities and to create fexible capacity, reduce costs and
share risks to solve complex R&D problems together. Intellectual property protection and
data package exclusivity are necessary to support investments in pharmaceutical R&D. Its
also important that we provide our cutting-edge drug hunters and developers with the right
scientifc capabilities, supported by organizational empowerment to encourage transparent
risk-taking and the willingness to play to win even in a difcult and uncertain environment
rather than playing safe not to lose. Science has been, and will continue to be, a foundation
upon which innovation driven pharmaceutical companies succeed or fail. We believe that
pharma mega mergers risk sequestering innovation, create distraction and may destroy
unique capabilities that have been built up for years. Instead, we believe that optimal critical
mass capability, continuous adoption of new science and technology with a long-term
approach is the way forward to truly supporting innovation.
In this reprint collection, we provide a compilation of some of the papers which describe
issues and solutions for innovation and create a dialogue around which efective strategies
can be built. Unmet medical need has an eternal horizon and the ultimate reward for our
scientists is to invent meaningful medicines that help humans live longer, active and healthier
lives. That is the ultimate reward for all of us who do this important work and contribute
positively to solving societys healthcare and economic challenges.
Jan M. Lundberg, Ph.D.
Executive Vice President, Science and Technology
President, Lilly Research Laboratories
This supplement is published by Nature Publishing Group on behalf of
Eli Lilly and Company. All content has been chosen by Eli Lilly and Company.
2 How to improve R&D productivity: the
pharmaceutical industrys grand challenge. Paul,
S.M. et al. Nature Reviews Drug Discovery 9, 203-214
(2010); doi: 10.1038/nrd3078
14 Network pharmacology: the next paradigm in drug
discovery. Hopkins, A.L. Nature Chemical Biology 4,
682-690 (2008); doi: 10.1038/nchembio.118
23 How were new medicines discovered? Swinney,
D.C. et al. Nature Reviews Drug Discovery 10, 507-519
(2010); doi: 10.1038/nrd3480
37 Development trends for human monoclonal
antibody therapeutics. Nelson, A.L. et al. Nature
Reviews Drug Discovery 9, 767-774 (2010); doi:
10.1038/nrd3229
45 Bridging the efcacyefectiveness gap: a
regulators perspective on addressing variability
of drug response. Eichler, HG. et al. Nature Reviews
Drug Discovery 10, 495-506 (2011); doi: 10.1038/
nrd3501
57 Innovative drug R&D in China. Qi, J. et al. Nature
10.1038/nrd3435
59 The impact of mergers on pharmaceutical R&D.
LaMattina, J.L. Nature Reviews Drug Discovery 10,
559-560 (2011); doi: 10.1038/nrd3514
61 Crowd sourcing in drug discovery. Lessl, M. et al.
Nature Reviews Drug Discovery 10, 241-242 (2011);
doi: 10.1038/nrd3412
63 The future of drug development: advancing clinical
trial design. Orlof, J. et al. Nature Reviews Drug
Discovery 8, 949-957 (2009); doi: 10.1038/nrd3025
73 The case for entrepreneurship in R&D in the
pharmaceutical industry. Douglas, F.L. Nature
10.1038/nrd3230
2 | OCTOBER 2011 | THE FUTURE OF DRUG INNOVATION | NATURE REPRINT COLLECTION www.nature.com/reprintcollections/lilly/drug-innovation
innovative and first-in-class NMEs have remained
stable at about 56 per year. however, the number of
potential revenue-generating drugs (innovative or
other wise) as a percentage of R&D expenditures has
undeniably fallen sharply.
with an estimated $50 billion in collective annual
R&D spending by the large pharmaceutical companies,
and appropriate allocation over time to the successful
discovery and development of NMEs, the average cost
for these companies to bring an NME to market is now
estimated to be approximately $1.8 billion (see below for
details underlying this estimate), and is rising rapidly.
Moreover, there is little evidence that the average costs
of successfully launching an NME vary significantly
between large pharmaceutical or small biotechnology
companies
10,11
.
Although R&D productivity has been declining
for a number of years
2
, the unprecedented combina-
tion of reduced R&D output in the form of success-
fully launched truly innovative NMEs, coupled with
diminishing market exclusivity for recently launched
new medicines and the huge loss of revenues owing to
generic competition over the next decade, suggest that
we may be moving closer to a pharmaceutical ice age
and the potential extinction of the industry, at least as it
exists today
12,13
. Although this might be welcomed by the
industrys critics, the impact on the health and well-being
of patients owing to delayed or even lost opportunities
to introduce the next generation of innovative medicines
could be devastating. In this regard, we underscore the
findings of lichtenberg
14
on the effects of medical inno-
vation (including controls for the impact of obesity and
income), which indicate that ~40% of the 2-year increase
in life expectancy measured from 19862000 can be
attributed to the introduction and use of new drugs. It
took approximately 3 years for NME launches to have
their maximal impact on longevity this effect was
not observed for non-NME (older) drugs. One can only
speculate as to the impact on longevity and quality of life
that new drugs now in clinical development for cancer
and Alzheimers disease might have. without these new
medicines, and given the rise in diseases such as diabetes
and childhood obesity, it is possible that life expectancy
may actually decrease over time
15
.
Among all the challenges faced by the pharmaceutical
industry, we argue that improving R&D productivity
remains the most important. The environmental factors
that are reducing the industrys profitability can only
be mitigated by substantially and sustainably increas-
ing the number and quality of innovative, as well as
cost-effective, new medicines; but only if accomplished
at reasonable R&D costs. So, the key questions are
where, how and by how much can R&D productivity
be improved? here, we present a detailed analysis of
R&D productivity by first defining and modelling the
essential elements of contemporary drug discovery
and development that account for the current cost of
a new medicine, and discuss the rate-limiting steps of
the R&D process that are contributing to reduced R&D
productivity. we then propose, and illustrate, ways to
improve these factors.
How do we define R&D productivity?
R&D productivity can be simply defined as the relation-
ship between the value (medical and commercial) created
by a new medicine (considered here to be an NME)
and the investments required to generate that medicine.
however, R&D productivity can in our view best be
elaborated in two important dimensions: inputs leading
to outputs, or R&D efficiency; and outputs leading to
outcomes, or R&D effectiveness (FIG. 1).
R&D efficiency represents the ability of an R&D
system to translate inputs (for example, ideas, invest-
ments, effort) into defined outputs (for example, inter-
nal milestones that represent resolved uncertainty for
a given project or product launches), generally over a
defined period of time. If launching (gaining regulatory
approval and commercializing) an NME is the desired
output, how can this be achieved with greater efficiency
(that is, at a lower cost)?
R&D effectiveness can be defined as the ability of the
R&D system to produce outputs with certain intended
and desired qualities (for example, medical value to
patients, physicians and payers, and substantial com-
mercial value). Thus, R&D productivity can be viewed
as an aggregate representation of both the efficiency and
effectiveness of the drug discovery and development
process; the goal of a highly productive R&D system is
to efficiently translate inputs into the most desired and
valuable outputs. For a more detailed description of these
definitions, see Supplementary information S1 (box).
with this definition of R&D productivity in mind, we
have further adapted a productivity relationship or
pharmaceutical value equation, which includes the key
elements that determine both the efficiency and effec-
tiveness of the drug discovery and development process
for any given pipeline (see equation 1).
Nature Reviews | Drug Discovery
P (1)
WIP p(TS) V
CT C
R&D productivity (P) can be viewed as a function of the
elements comprising the numerator the amount of
scientific and clinical research being conducted simul-
taneously, designated here as the work in process (WIP),
the probability of technical success (p(TS)) and the value
(V) divided by the elements in the denominator, the
cycle time (CT) and cost (C). Each of these parameters
can be conceptualized and analyzed on a per project
basis (for example, a single drug candidate or WIP = 1)
or collectively as a larger portfolio or pipeline of projects
or drug candidates. In general, increasing the numerator
relative to the denominator will increase productivity
and vice versa. Thus, if one could increase the p(TS)
(that is, reduce attrition) for any given drug candidate
or ideally for a portfolio of drug candidates at a given
phase of development, P would increase accordingly.
Similarly, for any given level of R&D investment, sub-
stantially reducing CT or lowering C (such as unit costs)
would increase P.
however, most of the elements comprising equa-
tion 1 are inextricably linked to one another and changing
one element can often adversely or beneficially affect
204 | MARch 2010 | vOlUME 9 www.nature.com/reviews/drugdisc
nrd_3078_mar10.indd 204 15/2/10 11:08:15
AnAlysi s
The pharmaceutical industry is facing unprecedented
challenges to its business model. Experienced observers
and industry analysts have even predicted its imminent
demise
13
. Over the past decade, serious concerns about
the industrys integrity and transparency for example,
around drug safety and efficacy have been raised,
compromising the industrys image, and resulting in
increased regulatory scrutiny
4,5
. This erosion in confi-
dence in the industry and its products has resonated
poorly with patients, health-care professionals, payers
and shareholders. Indeed, the industrys price/earnings
ratio, a measure of the current valuation of the industry,
has decreased below that of the S&P 500 index and has
remained more or less flat, as have share prices for the
past 7 years.
The industrys profitability and growth prospects
are also under pressure as healthcare budgets become
increasingly strained. Generic drugs, although clearly
helping to keep drug prices in check, are currently
approaching 70% of all prescriptions written in the
United States
6
. Moreover, key patent expirations between
20102014 have been estimated to put more than US$209
billion in annual drug sales at risk, resulting in $113
billion of sales being lost to generic substitution
7
. Indeed,
for every dollar lost in declining product revenues due
to patent expirations by 2012, it has been estimated
that large-cap pharmaceutical companies will only be
able to replace on average 26 cents with new product
revenues
8
.
Simply stated, without a dramatic increase in R&D
productivity, todays pharmaceutical industry cannot
sustain sufficient innovation to replace the loss of rev-
enues due to patent expirations for successful products.
A key aspect of this problem is the decreasing number
of truly innovative new medicines approved by the
US Food and Drug Administration (FDA) and other
major regulatory bodies around the world over the
past 5 years (in which 50% fewer new molecular entities
(NMEs) were approved compared with the previous
5 years)
9
. In 2007, for example, only 19 NMEs (including
biologics) were approved by the FDA, the fewest
number of NMEs approved since 1983, and the number
rose only slightly to 21 in 2008. Of the 21 new drugs
approved by the FDA in 2008, only 6 were developed by
the 15 largest pharmaceutical companies and only 29%
would be considered first-in-class medicines. In 2009,
24 new drugs were approved, 10 of which were devel-
oped by large pharmaceutical companies and only 17%
of which could be considered first-in-class. Some have
argued that the number of approved mechanistically
Lilly Research Laboratories,
Eli Lilly and Company,
Lilly Corporate Center,
Indianapolis, Indiana
46285, USA.
Correspondence to: S.M.P.
e-mail:
smpaulmd@gmail.com
doi:10.1038/nrd3078
Published online
19 February 2010
New molecular entity
(NME). A medication
containing an active ingredient
that has not been previously
approved for marketing in any
form in the United States. NME
is conventionally used to refer
only to small-molecule drugs,
but in this article we use the
term as a shorthand to refer to
both new chemical entities and
new biologic entities.
How to improve R&D productivity:
the pharmaceutical industrys grand
challenge
Steven M. Paul, Daniel S. Mytelka, Christopher T. Dunwiddie, Charles C. Persinger,
Bernard H. Munos, Stacy R. Lindborg and Aaron L. Schacht
Abstract | The pharmaceutical industry is under growing pressure from a range of
environmental issues, including major losses of revenue owing to patent expirations,
increasingly cost-constrained healthcare systems and more demanding regulatory
requirements. In our view, the key to tackling the challenges such issues pose to both the
future viability of the pharmaceutical industry and advances in healthcare is to substantially
increase the number and quality of innovative, cost-effective new medicines, without
incurring unsustainable R&D costs. However, it is widely acknowledged that trends in
industry R&D productivity have been moving in the opposite direction for a number of years.
Here, we present a detailed analysis based on comprehensive, recent, industry-wide data
to identify the relative contributions of each of the steps in the drug discovery and
development process to overall R&D productivity. We then propose specific strategies
that could have the most substantial impact in improving R&D productivity.
AnAlysIs
NATURE REvIEwS | Drug Discovery vOlUME 9 | MARch 2010 | 203
nrd_3078_mar10.indd 203 15/2/10 11:08:13
First published in Nature Reviews Drug Discovery 9, 203-214 (March 2010) | doi:10.1038/nrd3078
innovative and first-in-class NMEs have remained
stable at about 56 per year. however, the number of
potential revenue-generating drugs (innovative or
other wise) as a percentage of R&D expenditures has
undeniably fallen sharply.
with an estimated $50 billion in collective annual
R&D spending by the large pharmaceutical companies,
and appropriate allocation over time to the successful
discovery and development of NMEs, the average cost
for these companies to bring an NME to market is now
estimated to be approximately $1.8 billion (see below for
details underlying this estimate), and is rising rapidly.
Moreover, there is little evidence that the average costs
of successfully launching an NME vary significantly
between large pharmaceutical or small biotechnology
companies
10,11
.
Although R&D productivity has been declining
for a number of years
2
, the unprecedented combina-
tion of reduced R&D output in the form of success-
fully launched truly innovative NMEs, coupled with
diminishing market exclusivity for recently launched
new medicines and the huge loss of revenues owing to
generic competition over the next decade, suggest that
we may be moving closer to a pharmaceutical ice age
and the potential extinction of the industry, at least as it
exists today
12,13
. Although this might be welcomed by the
industrys critics, the impact on the health and well-being
of patients owing to delayed or even lost opportunities
to introduce the next generation of innovative medicines
could be devastating. In this regard, we underscore the
findings of lichtenberg
14
on the effects of medical inno-
vation (including controls for the impact of obesity and
income), which indicate that ~40% of the 2-year increase
in life expectancy measured from 19862000 can be
attributed to the introduction and use of new drugs. It
took approximately 3 years for NME launches to have
their maximal impact on longevity this effect was
not observed for non-NME (older) drugs. One can only
speculate as to the impact on longevity and quality of life
that new drugs now in clinical development for cancer
and Alzheimers disease might have. without these new
medicines, and given the rise in diseases such as diabetes
and childhood obesity, it is possible that life expectancy
may actually decrease over time
15
.
Among all the challenges faced by the pharmaceutical
industry, we argue that improving R&D productivity
remains the most important. The environmental factors
that are reducing the industrys profitability can only
be mitigated by substantially and sustainably increas-
ing the number and quality of innovative, as well as
cost-effective, new medicines; but only if accomplished
at reasonable R&D costs. So, the key questions are
where, how and by how much can R&D productivity
be improved? here, we present a detailed analysis of
R&D productivity by first defining and modelling the
essential elements of contemporary drug discovery
and development that account for the current cost of
a new medicine, and discuss the rate-limiting steps of
the R&D process that are contributing to reduced R&D
productivity. we then propose, and illustrate, ways to
improve these factors.
How do we define R&D productivity?
R&D productivity can be simply defined as the relation-
ship between the value (medical and commercial) created
by a new medicine (considered here to be an NME)
and the investments required to generate that medicine.
however, R&D productivity can in our view best be
elaborated in two important dimensions: inputs leading
to outputs, or R&D efficiency; and outputs leading to
outcomes, or R&D effectiveness (FIG. 1).
R&D efficiency represents the ability of an R&D
system to translate inputs (for example, ideas, invest-
ments, effort) into defined outputs (for example, inter-
nal milestones that represent resolved uncertainty for
a given project or product launches), generally over a
defined period of time. If launching (gaining regulatory
approval and commercializing) an NME is the desired
output, how can this be achieved with greater efficiency
(that is, at a lower cost)?
R&D effectiveness can be defined as the ability of the
R&D system to produce outputs with certain intended
and desired qualities (for example, medical value to
patients, physicians and payers, and substantial com-
mercial value). Thus, R&D productivity can be viewed
as an aggregate representation of both the efficiency and
effectiveness of the drug discovery and development
process; the goal of a highly productive R&D system is
to efficiently translate inputs into the most desired and
valuable outputs. For a more detailed description of these
definitions, see Supplementary information S1 (box).
with this definition of R&D productivity in mind, we
have further adapted a productivity relationship or
pharmaceutical value equation, which includes the key
elements that determine both the efficiency and effec-
tiveness of the drug discovery and development process
for any given pipeline (see equation 1).
P (1)
WIP p(TS) V
CT C
R&D productivity (P) can be viewed as a function of the
elements comprising the numerator the amount of
scientific and clinical research being conducted simul-
taneously, designated here as the work in process (WIP),
the probability of technical success (p(TS)) and the value
(V) divided by the elements in the denominator, the
cycle time (CT) and cost (C). Each of these parameters
can be conceptualized and analyzed on a per project
basis (for example, a single drug candidate or WIP = 1)
or collectively as a larger portfolio or pipeline of projects
or drug candidates. In general, increasing the numerator
relative to the denominator will increase productivity
and vice versa. Thus, if one could increase the p(TS)
(that is, reduce attrition) for any given drug candidate
or ideally for a portfolio of drug candidates at a given
phase of development, P would increase accordingly.
Similarly, for any given level of R&D investment, sub-
stantially reducing CT or lowering C (such as unit costs)
would increase P.
however, most of the elements comprising equa-
tion 1 are inextricably linked to one another and changing
one element can often adversely or beneficially affect
nrd_3078_mar10.indd 204 15/2/10 11:08:15
AnAlysi s
The pharmaceutical industry is facing unprecedented
challenges to its business model. Experienced observers
and industry analysts have even predicted its imminent
demise
13
. Over the past decade, serious concerns about
the industrys integrity and transparency for example,
around drug safety and efficacy have been raised,
compromising the industrys image, and resulting in
increased regulatory scrutiny
4,5
. This erosion in confi-
dence in the industry and its products has resonated
poorly with patients, health-care professionals, payers
and shareholders. Indeed, the industrys price/earnings
ratio, a measure of the current valuation of the industry,
has decreased below that of the S&P 500 index and has
remained more or less flat, as have share prices for the
past 7 years.
The industrys profitability and growth prospects
are also under pressure as healthcare budgets become
increasingly strained. Generic drugs, although clearly
helping to keep drug prices in check, are currently
approaching 70% of all prescriptions written in the
United States
6
. Moreover, key patent expirations between
20102014 have been estimated to put more than US$209
billion in annual drug sales at risk, resulting in $113
billion of sales being lost to generic substitution
7
. Indeed,
for every dollar lost in declining product revenues due
to patent expirations by 2012, it has been estimated
that large-cap pharmaceutical companies will only be
able to replace on average 26 cents with new product
revenues
8
.
Simply stated, without a dramatic increase in R&D
productivity, todays pharmaceutical industry cannot
sustain sufficient innovation to replace the loss of rev-
enues due to patent expirations for successful products.
A key aspect of this problem is the decreasing number
of truly innovative new medicines approved by the
US Food and Drug Administration (FDA) and other
major regulatory bodies around the world over the
past 5 years (in which 50% fewer new molecular entities
(NMEs) were approved compared with the previous
5 years)
9
. In 2007, for example, only 19 NMEs (including
biologics) were approved by the FDA, the fewest
number of NMEs approved since 1983, and the number
rose only slightly to 21 in 2008. Of the 21 new drugs
approved by the FDA in 2008, only 6 were developed by
the 15 largest pharmaceutical companies and only 29%
would be considered first-in-class medicines. In 2009,
24 new drugs were approved, 10 of which were devel-
oped by large pharmaceutical companies and only 17%
of which could be considered first-in-class. Some have
argued that the number of approved mechanistically
Lilly Research Laboratories,
Eli Lilly and Company,
Lilly Corporate Center,
Indianapolis, Indiana
46285, USA.
Correspondence to: S.M.P.
e-mail:
smpaulmd@gmail.com
doi:10.1038/nrd3078
Published online
19 February 2010
New molecular entity
(NME). A medication
form in the United States. NME
is conventionally used to refer
only to small-molecule drugs,
but in this article we use the
term as a shorthand to refer to
both new chemical entities and
new biologic entities.
How to improve R&D productivity:
the pharmaceutical industrys grand
challenge
Steven M. Paul, Daniel S. Mytelka, Christopher T. Dunwiddie, Charles C. Persinger,
Bernard H. Munos, Stacy R. Lindborg and Aaron L. Schacht
Abstract | The pharmaceutical industry is under growing pressure from a range of
environmental issues, including major losses of revenue owing to patent expirations,
increasingly cost-constrained healthcare systems and more demanding regulatory
requirements. In our view, the key to tackling the challenges such issues pose to both the
future viability of the pharmaceutical industry and advances in healthcare is to substantially
increase the number and quality of innovative, cost-effective new medicines, without
incurring unsustainable R&D costs. However, it is widely acknowledged that trends in
industry R&D productivity have been moving in the opposite direction for a number of years.
Here, we present a detailed analysis based on comprehensive, recent, industry-wide data
to identify the relative contributions of each of the steps in the drug discovery and
development process to overall R&D productivity. We then propose specific strategies
that could have the most substantial impact in improving R&D productivity.
AnAlysIs
nrd_3078_mar10.indd 203 15/2/10 11:08:13
Launch
p(TS)
WIP needed for 1 launch
Cost per WIP per Phase
Cycle time (years)
Cost per launch (out of pocket)
% Total cost per NME
Cost of capital
Cost per launch (capitalized)
Target-to-hit
80%
24.3
$1
1.0
$24
3%
11%
$94
Hit-to-lead
75%
19.4
$2.5
1.5
$49
6%
$166
Lead
optimization
85%
14.6
$10
2.0
$146
17%
$414
Phase I
54%
8.6
$15
1.5
$128
15%
$273
Phase II
34%
4.6
$40
2.5
$185
21%
$319
Phase III
70%
1.6
$150
2.5
$235
27%
$314
Submission
to launch
91%
1.1
$40
1.5
$44
5%
$48
1
$873
$1,778
Preclinical
69%
12.4
$5
1.0
$62
7%
$150
Discovery Development
20 and 15 respectively (FIG. 2). we will discuss the need
for sufficient discovery investments and output (WIP)
to achieve the level of drug candidates necessary below.
In this model, in the absence of sufficient acquisition
of drug candidates, especially late-phase compounds,
achieved as one-off in-license deals or through mergers
and acquisitions (M&A), most companies are simply
unable to achieve (or afford) the numbers of compounds
distributed across the phases of discovery and develop-
ment they require to achieve their goals for new NMEs
launched without a substantial increase in productivity.
Encouragingly, recent benchmark data on Phase I
WIP across the industry indicate that most companies
have begun to substantially increase investments in the
earlier stages of drug discovery; this is reflected by the
number of candidates entering Phase I trials, which
has increased significantly
9,17,18
. however, based on the
benchmark data, for most companies, the number of
NMEs entering clinical development and progressing
to Phase II and III are still insufficient to achieve 25
launches per year
9
; this reflects many years of operating
at WIP levels below what would be required in the ear-
lier stages of drug discovery and development. Thus,
inevitable pipeline gaps will arise (as they have) and
given the CT of the process (FIG. 2), such gaps cannot be
filled quickly through traditional means.
Finally, we suggest that based on this model, many
companies would find that their R&D operating
expenses are not appropriately distributed across the
various phases of drug discovery and development. Too
many resources are often applied to late-stage develop-
ment of drug candidates with relatively low p(TS) and/
or post-launch support of marketed products. This may
be the root cause of the current drought of new medi-
cines and the business challenges most companies are
experiencing.
Key areas for improving R&D productivity
Using our model (FIG. 2, Supplementary information S2
(box)) and starting from a baseline value for the estimated
capitalized cost of a single NME of ~$1.78 billion, we can
investigate which parameters contributing to this cost are
the most important. To achieve this, we have varied the
parameters p(TS), CT and C for different phases of the
overall process across a realistic range of possibilities
(reasonable estimates of industry highs and lows for each
parameter) to identify parameters for which changes
would have the greatest impact on R&D efficiency, and
the extent of the impact in each case (FIG. 3).
As is evident from FIG. 3, attrition defined as
1 p(TS) in the clinical phases of development (espe-
cially Phase II and III) remains the most important
Figure 2 | r&D model yielding costs to successfully discover and develop a single new molecular entity. The model
defines the distinct phases of drug discovery and development from the initial stage of target-to-hit to the final stage, launch.
The model is based on a set of industry-appropriate R&D assumptions (industry benchmarks and data from Eli lilly and
Company) defining the performance of the R&D process at each stage of development (see supplementary information s2
(box) for details). R&D parameters include: the probability of successful transition from one stage to the next (p(Ts)), the phase
cost for each project, the cycle time required to progress through each stage of development and the cost of capital,
reflecting the returns required by shareholders to use their money during the lengthy R&D process. With these inputs (darker
shaded boxes), the model calculates the number of assets (work in process, WIP) needed in each stage of development to
achieve one new molecular entity (nME) launch. Based on the assumptions for success rate, cycle time and cost, the model
further calculates the out of pocket cost per phase as well as the total cost to achieve one nME launch per year (Us$873
million). lighter shaded boxes show calculated values based on assumed inputs. Capitalizing the cost, to account for the cost
of capital during this period of over 13 years, yields a capitalized cost of $1,778 million per nME launch. It is important to
note that this model does not include investments for exploratory discovery research, post-launch expenses or overheads
(that is, salaries for employees not engaged in R&D activities but necessary to support the organization).
nrd_3078_mar10.indd 206 15/2/10 11:08:16
AnAlysi s
Inputs
R&D efficiency
More affordable drugs
via less costly R&D
R&D effectiveness
More value for the patient
via innovative drugs with
high-quality information
Outputs Outcomes
Cost per launch Value per launch
Capitalized cost
This is the out-of-pocket cost
corrected for cost of capital,
and is the standard accounting
treatment for long-term
investments. It recognizes the
fact that investors require a
return on research investments
that reflects alternative
potential uses of their
investment. So, the capitalized
cost per drug launch increases
out-of-pocket costs by the cost
of capital for every year from
expenditure to launch.
Out-of-pocket cost
This is the total cost required
to expect one drug launch,
taking into account attrition,
but not the cost of capital.
Cost of capital
This is the annual rate of return
expected by investors based
on the level of risk of the
investment.
another. For example, as discussed below, having suf-
ficient pipeline WIP (by phase of development) is
crucial given the substantial phase-specific attrition
rates. however, increasing WIP (especially late-phase
WIP) alone will undoubtedly increase C and may also
increase CT, which could further reduce P and diminish
productivity.
Finally, although carrying out definitive health out-
come studies on late-stage compounds before approval
is often highly desirable and increasingly necessary to
unequivocally demonstrate value (V) for reimbursement
purposes, such studies can substantially increase CT and
C, thus also diminishing P. Nevertheless, such studies
will also increase V, potentially offsetting any decrease,
or even increasing, P.
A model of R&D productivity
To inform efforts to increase R&D productivity (P), the
key questions include: which of the associated elements
have the greatest impact; how might they be improved;
and by what magnitude? To help address these questions,
we have built an economic model of drug discovery and
development which, using industry-appropriate assump-
tions, provides the basis for our estimate that the fully
capitalized cost of an average NME developed by a typi-
cal large pharmaceutical company is currently ~$1.8
billion) (see Supplementary information S2 (box) for
details). The model has been constructed using recently
available R&D performance productivity data from a
group of 13 large pharmaceutical companies, provided
by the Pharmaceutical Benchmarking Forum (PBF)
16

(see Supplementary information S3 (box)), as well as
our own internal data, to closely approximate the key
elements of our productivity relationship that underlie
R&D efficiency C, WIP, CT and p(TS) for each
phase of discovery and development (FIG. 2).
we recognize that the estimated cost per NME is
highly dependent on a number of economic or financial
assumptions. consequently, for our estimated cost of an
NME we show both out of pocket and capitalized costs
using a cost of capital of 11% (FIG. 2). Our estimate repre-
sents molecule only costs and does not include the costs
of exploratory discovery research (target identification
and validation) or other non-molecule costs (which
include overheads, such as salaries for employees that
are not engaged in research and development activities
but that are otherwise necessary to support the R&D
organization; these represent approximately 2030% of
total costs). we discuss comparisons of our estimates
with other reported estimates in Supplementary infor-
mation S2 (box). however, for modelling purposes, the
exact cost per NME is not crucial as long as our assump-
tions for each parameter in our model are consistent and
represent reasonable estimates. Each R&D organization
can (and should) build a similar model based on their
own data, which may vary from company to company.
The exact output of the model the desired number
of new launches (and the estimated commercial value
per launch) will depend on business aspirations, ther-
apeutic focus and absolute level of R&D investments of a
given company. Nonetheless, based on our model, a few
key observations can be made.
First, clinical development (Phases IIII) accounts for
approximately 63% of the costs for each NME launched
(53% from Phase II to launch), and preclinical drug dis-
covery accounts for 32%. however, this represents an
underestimate of the costs for drug discovery, as we have
excluded from our model the earliest phase of discovery
research; that is, that prior to target selection. This is
because the research required to identify and validate
a given target is highly variable, making the underlying
parameters difficult to quantify. however, target selec-
tion may well be one of the most important determinants
of attrition (p(TS)) and thus overall R&D productivity
(discussed below).
Second, based on realistic and current assumptions
on C, CT, p(TS) and WIP, only 8% of NMEs will success-
fully make it from the point of candidate selection (pre-
clinical stage) to launch

(FIG. 2). It has been suggested that
new biologic drugs have a higher probability of launch
than small-molecule drugs
9,11
. For the purposes of our
model, we have used 7% for small-molecule drugs and
11% for biologics.
Third, the process of discovering and developing an
NME on average required approximately 13.5 years (CT)
in 2007 (yearly averages ranged from 11.4 to 13.5 using
the PBF study data across 20002007). This includes
regulatory review but not the time it takes to fully identify
and validate a drug target
16
.
Fourth, based on our model, the number of mol-
ecules entering clinical development every year must be
approximately 9 (or 11 if all small molecules) to yield a
single NME launch per year. Most large companies aspire
for 25 launches per year and therefore 1845 Phase I
starts (and resulting WIP) would be required annually.
however, such numbers are rarely, if ever, achieved even
in very large companies. If sustained over several years,
this WIP deficit will result in a substantial pipeline gap. If
it takes approximately 9 Phase I drug candidates annually
to launch 1 NME per year and if these derive exclusively
from a given companys internal discovery efforts, then
the number of discovery projects (WIP) from target-to-
hit, hit-to-lead and lead optimization is approximately 25,
Figure 1 | Dimensions of r&D productivity. To improve
R&D productivity, it is crucial to understand the
interdependencies between inputs (for example, R&D
investments), output (for example, new molecular entity
launches) and outcomes (for example, valued outcomes
for patients). This figure outlines the key dimensions of
R&D productivity and the goals tied to R&D efficiency
and effectiveness. An effective R&D productivity strategy
must encompass both of these components. Value will
be created by delivering innovative products with
high-quality information.
nrd_3078_mar10.indd 205 15/2/10 11:08:16
AnAlysi s
Launch
p(TS)
WIP needed for 1 launch
Cost per WIP per Phase
Cycle time (years)
Cost per launch (out of pocket)
% Total cost per NME
Cost of capital
Cost per launch (capitalized)
Target-to-hit
80%
24.3
$1
1.0
$24
3%
11%
$94
Hit-to-lead
75%
19.4
$2.5
1.5
$49
6%
$166
Lead
optimization
85%
14.6
$10
2.0
$146
17%
$414
Phase I
54%
8.6
$15
1.5
$128
15%
$273
Phase II
34%
4.6
$40
2.5
$185
21%
$319
Phase III
70%
1.6
$150
2.5
$235
27%
$314
Submission
to launch
91%
1.1
$40
1.5
$44
5%
$48
1
$873
$1,778
Preclinical
69%
12.4
$5
1.0
$62
7%
$150
Discovery Development
20 and 15 respectively (FIG. 2). we will discuss the need
for sufficient discovery investments and output (WIP)
to achieve the level of drug candidates necessary below.
In this model, in the absence of sufficient acquisition
of drug candidates, especially late-phase compounds,
achieved as one-off in-license deals or through mergers
and acquisitions (M&A), most companies are simply
unable to achieve (or afford) the numbers of compounds
distributed across the phases of discovery and develop-
ment they require to achieve their goals for new NMEs
launched without a substantial increase in productivity.
Encouragingly, recent benchmark data on Phase I
WIP across the industry indicate that most companies
have begun to substantially increase investments in the
earlier stages of drug discovery; this is reflected by the
number of candidates entering Phase I trials, which
has increased significantly
9,17,18
. however, based on the
benchmark data, for most companies, the number of
NMEs entering clinical development and progressing
to Phase II and III are still insufficient to achieve 25
launches per year
9
; this reflects many years of operating
at WIP levels below what would be required in the ear-
lier stages of drug discovery and development. Thus,
inevitable pipeline gaps will arise (as they have) and
given the CT of the process (FIG. 2), such gaps cannot be
filled quickly through traditional means.
Finally, we suggest that based on this model, many
companies would find that their R&D operating
expenses are not appropriately distributed across the
various phases of drug discovery and development. Too
many resources are often applied to late-stage develop-
ment of drug candidates with relatively low p(TS) and/
or post-launch support of marketed products. This may
be the root cause of the current drought of new medi-
cines and the business challenges most companies are
experiencing.
Key areas for improving R&D productivity
Using our model (FIG. 2, Supplementary information S2
(box)) and starting from a baseline value for the estimated
capitalized cost of a single NME of ~$1.78 billion, we can
investigate which parameters contributing to this cost are
the most important. To achieve this, we have varied the
parameters p(TS), CT and C for different phases of the
overall process across a realistic range of possibilities
(reasonable estimates of industry highs and lows for each
parameter) to identify parameters for which changes
would have the greatest impact on R&D efficiency, and
the extent of the impact in each case (FIG. 3).
As is evident from FIG. 3, attrition defined as
1 p(TS) in the clinical phases of development (espe-
cially Phase II and III) remains the most important
Figure 2 | r&D model yielding costs to successfully discover and develop a single new molecular entity. The model
defines the distinct phases of drug discovery and development from the initial stage of target-to-hit to the final stage, launch.
The model is based on a set of industry-appropriate R&D assumptions (industry benchmarks and data from Eli lilly and
Company) defining the performance of the R&D process at each stage of development (see supplementary information s2
(box) for details). R&D parameters include: the probability of successful transition from one stage to the next (p(Ts)), the phase
cost for each project, the cycle time required to progress through each stage of development and the cost of capital,
reflecting the returns required by shareholders to use their money during the lengthy R&D process. With these inputs (darker
shaded boxes), the model calculates the number of assets (work in process, WIP) needed in each stage of development to
achieve one new molecular entity (nME) launch. Based on the assumptions for success rate, cycle time and cost, the model
further calculates the out of pocket cost per phase as well as the total cost to achieve one nME launch per year (Us$873
million). lighter shaded boxes show calculated values based on assumed inputs. Capitalizing the cost, to account for the cost
of capital during this period of over 13 years, yields a capitalized cost of $1,778 million per nME launch. It is important to
note that this model does not include investments for exploratory discovery research, post-launch expenses or overheads
(that is, salaries for employees not engaged in R&D activities but necessary to support the organization).
nrd_3078_mar10.indd 206 15/2/10 11:08:16
AnAlysi s
Inputs
R&D efficiency
More affordable drugs
via less costly R&D
R&D effectiveness
More value for the patient
via innovative drugs with
high-quality information
Outputs Outcomes
Cost per launch Value per launch
Capitalized cost
This is the out-of-pocket cost
corrected for cost of capital,
and is the standard accounting
treatment for long-term
investments. It recognizes the
fact that investors require a
return on research investments
that reflects alternative
potential uses of their
investment. So, the capitalized
cost per drug launch increases
out-of-pocket costs by the cost
of capital for every year from
expenditure to launch.
Out-of-pocket cost
This is the total cost required
to expect one drug launch,
taking into account attrition,
but not the cost of capital.
Cost of capital
This is the annual rate of return
expected by investors based
on the level of risk of the
investment.
another. For example, as discussed below, having suf-
ficient pipeline WIP (by phase of development) is
crucial given the substantial phase-specific attrition
rates. however, increasing WIP (especially late-phase
WIP) alone will undoubtedly increase C and may also
increase CT, which could further reduce P and diminish
productivity.
Finally, although carrying out definitive health out-
come studies on late-stage compounds before approval
is often highly desirable and increasingly necessary to
unequivocally demonstrate value (V) for reimbursement
purposes, such studies can substantially increase CT and
C, thus also diminishing P. Nevertheless, such studies
will also increase V, potentially offsetting any decrease,
or even increasing, P.
A model of R&D productivity
To inform efforts to increase R&D productivity (P), the
key questions include: which of the associated elements
have the greatest impact; how might they be improved;
and by what magnitude? To help address these questions,
we have built an economic model of drug discovery and
development which, using industry-appropriate assump-
tions, provides the basis for our estimate that the fully
capitalized cost of an average NME developed by a typi-
cal large pharmaceutical company is currently ~$1.8
billion) (see Supplementary information S2 (box) for
details). The model has been constructed using recently
available R&D performance productivity data from a
group of 13 large pharmaceutical companies, provided
by the Pharmaceutical Benchmarking Forum (PBF)
16

(see Supplementary information S3 (box)), as well as
our own internal data, to closely approximate the key
elements of our productivity relationship that underlie
R&D efficiency C, WIP, CT and p(TS) for each
phase of discovery and development (FIG. 2).
we recognize that the estimated cost per NME is
highly dependent on a number of economic or financial
assumptions. consequently, for our estimated cost of an
NME we show both out of pocket and capitalized costs
using a cost of capital of 11% (FIG. 2). Our estimate repre-
sents molecule only costs and does not include the costs
of exploratory discovery research (target identification
and validation) or other non-molecule costs (which
include overheads, such as salaries for employees that
are not engaged in research and development activities
but that are otherwise necessary to support the R&D
organization; these represent approximately 2030% of
total costs). we discuss comparisons of our estimates
with other reported estimates in Supplementary infor-
mation S2 (box). however, for modelling purposes, the
exact cost per NME is not crucial as long as our assump-
tions for each parameter in our model are consistent and
represent reasonable estimates. Each R&D organization
can (and should) build a similar model based on their
own data, which may vary from company to company.
The exact output of the model the desired number
of new launches (and the estimated commercial value
per launch) will depend on business aspirations, ther-
apeutic focus and absolute level of R&D investments of a
given company. Nonetheless, based on our model, a few
key observations can be made.
First, clinical development (Phases IIII) accounts for
approximately 63% of the costs for each NME launched
(53% from Phase II to launch), and preclinical drug dis-
covery accounts for 32%. however, this represents an
underestimate of the costs for drug discovery, as we have
excluded from our model the earliest phase of discovery
research; that is, that prior to target selection. This is
because the research required to identify and validate
a given target is highly variable, making the underlying
parameters difficult to quantify. however, target selec-
tion may well be one of the most important determinants
of attrition (p(TS)) and thus overall R&D productivity
(discussed below).
Second, based on realistic and current assumptions
on C, CT, p(TS) and WIP, only 8% of NMEs will success-
fully make it from the point of candidate selection (pre-
clinical stage) to launch

(FIG. 2). It has been suggested that
new biologic drugs have a higher probability of launch
than small-molecule drugs
9,11
. For the purposes of our
model, we have used 7% for small-molecule drugs and
11% for biologics.
Third, the process of discovering and developing an
NME on average required approximately 13.5 years (CT)
in 2007 (yearly averages ranged from 11.4 to 13.5 using
the PBF study data across 20002007). This includes
regulatory review but not the time it takes to fully identify
and validate a drug target
16
.
Fourth, based on our model, the number of mol-
ecules entering clinical development every year must be
approximately 9 (or 11 if all small molecules) to yield a
single NME launch per year. Most large companies aspire
for 25 launches per year and therefore 1845 Phase I
starts (and resulting WIP) would be required annually.
however, such numbers are rarely, if ever, achieved even
in very large companies. If sustained over several years,
this WIP deficit will result in a substantial pipeline gap. If
it takes approximately 9 Phase I drug candidates annually
to launch 1 NME per year and if these derive exclusively
from a given companys internal discovery efforts, then
the number of discovery projects (WIP) from target-to-
hit, hit-to-lead and lead optimization is approximately 25,
Figure 1 | Dimensions of r&D productivity. To improve
R&D productivity, it is crucial to understand the
interdependencies between inputs (for example, R&D
investments), output (for example, new molecular entity
launches) and outcomes (for example, valued outcomes
for patients). This figure outlines the key dimensions of
R&D productivity and the goals tied to R&D efficiency
and effectiveness. An effective R&D productivity strategy
must encompass both of these components. Value will
be created by delivering innovative products with
high-quality information.
nrd_3078_mar10.indd 205 15/2/10 11:08:16
AnAlysi s
P
h
a
s
e

I

e
n
t
r
i
e
s

(
F
H
D
s
)
20
18
16
14
12
10
8
2
4
6
0
15 20 25 30 35 40 45 50 55 60 65
Phase II p(TS) (%)
25%
50%
Phase III
p(TS) (%)
50%
60%
70%
80%
90%
Estimate used
in the model
trials each year to yield 1 NME launch. This number
decreases proportionately as the rates of Phase II and
III attrition decline (FIG. 4). Increasing Phase II and III
p(TS) to 50% and 80%, respectively, reduces the number
of Phase I entries needed per year by almost two thirds.
In other words, and everything else being equal, the
same overall R&D investment should yield 23 times
the number of NMEs at these lower attrition rates.
however, just increasing WIP alone without having
sufficient development capacity will probably have
a highly deleterious effect on CT. littles law
22
, which
relates the throughput or flow of a given project to WIP
and CT, posits that too much WIP will result in increased
CT, especially if development resources become rate-
limiting. The ability and need to accurately estimate the
development resources required (capacity management)
and to carefully and optimally balance WIP throughout
the phases of drug discovery and development must be
emphasized (see Bunch & Schacht
23
for a discussion on
capacity management). In the absence of a substantial
reduction in attrition, pharmaceutical companies must
find more affordable means to increase early-stage WIP
and to expeditiously advance these drug candidates
through the development stages.
how can pharmaceutical companies substantially
increase their pipeline WIP without dramatically
increasing C? First, there must be sufficient WIP in the
early stages of drug discovery, and especially early drug
development (as outlined above). Funding these early-
stage (Phase I and II) compounds, especially in the num-
bers indicated by our model, must come in part from
reducing investments in late-stage development, ideally
by redirecting resources from molecules destined to fail
in Phase III (or even Phase Iv). Given the C and CT of
a single Phase III unit of WIP ($150 million), almost 10
Phase I molecules ($15 million) can be developed for the
same cost, ideally through to proof-of-concept (POc;
see discussion of p(TS) below). Reducing late-phase
attrition through early POc studies (ideally in Phase I)
is therefore crucial to implement this partial solution.
The resources (C) saved by lowering Phase III attrition,
however, must be redirected to fund sufficient discovery
and Phase I/II WIP. Most importantly, advancement into
Phase III should be pursued only for those compounds
with established efficacy (ideally POc in Phase I and
confirmed in Phase II) and a well-defined margin of
safety. Ideally, attrition in Phase III should be due pri-
marily to the emergence of relatively rare and unforeseen
adverse events. Thus, the key is to have sufficient WIP
in the early phases of clinical development to effectively
triage and select molecules that will have a higher p(TS)
in late-stage development.
The question of how to affordably increase WIP,
p(TS) and V without substantially increasing C or
increasing CT due to capacity constraints and lack of
focus is in our view paramount to improve R&D pro-
ductivity. This could be accomplished by transforming
the R&D enterprise from one that is predominantly
owned, operated and fully controlled by a given com-
pany (Fully Integrated Pharmaceutical company or
FIPco) to one that is highly networked, partnered and
leveraged (Fully Integrated Pharmaceutical Network or
FIPNet). Traditionally, large pharmaceutical companies
have pursued the discovery, development, manufacture
and commercialization of their medicines largely by
owning and controlling each component. In part, past
reliance on the FIPco model was as much a necessity
as a choice. Today, however, the opportunity to partner
virtually all elements of R&D through a coordinated and
global network or FIPNet could (if effectively managed)
substantially improve R&D productivity by affordably
enhancing the pipeline from early discovery through to
launch. A FIPNet will theoretically allow greater access
to intellectual property, molecules, capabilities, capital,
knowledge and, of course, talent
2426
. Thus, operated as
a FIPNet, a given R&D organization will be able to play
bigger than its size and better leverage its resources to
increase WIP across the pipeline, as long as the costs
of managing the network do not become prohibitive.
Although a full discussion of such a R&D FIPNet is
beyond the scope of this article, we would emphasize
that it includes many types of partnerships ranging from
function-based outsourcing (for example, toxicology or
clinical development), to lower C and reduce CT in many
cases, to molecule-based risk-sharing partnerships and
even partial ownership or equity investments in smaller
companies or joint ventures with larger companies.
A successfully operated R&D FIPNet should aim to bet-
ter leverage and proportionally decrease C to affordably
increase pipeline WIP, while simultaneously mitigating
financial and technical risks.
Much of the discussion on WIP so far has focused on
ensuring that there is sufficient WIP to increase NME
approvals while addressing the inter-relatedness of the
Figure 4 | effect of Phase ii and iii probability of technical success on the number
of Phase i entries required for one successful launch of a new molecular entity.
This analysis shows the number of Phase I entries (first human dose; FHDs) annually
required to achieve one new molecular entity (nME) launch per year as a result of
modelling baseline assumptions of the probability of technical success (p(Ts)) for the
stages of Phase I and submission-to-launch (54% and 91% respectively) over a range of
p(Ts) for Phase II and Phase III. Each curve represents a different assumption for the
Phase III p(Ts) over the range of 50% to 90%, and the x axis represents varying p(Ts) for
Phase II. The number of Phase I entries (FHDs) annually needed to produce one nME
launch per year can be viewed on the y axis for any combination of Phase III (individual
curve) and Phase II (x axis) p(Ts). For example, at a 70% Phase III p(Ts) (black curve) and
a 35% Phase II p(Ts) (on x axis), the required number of Phase I entries is about 8.5.
nrd_3078_mar10.indd 208 15/2/10 11:08:18
AnAlysi s
p(TS): Phase II
p(TS): Phase III
Cost: lead optimization
Cycle time: Phase III
p(TS): Phase I
p(TS): submission to launch
Cycle time: Phase II
Cost: Phase II
Cost: Phase III
Cycle time: submission to launch
Cost: Phase I
p(TS): preclinical
Cost: hit-to-lead
p(TS): lead optimization
Cycle time: Phase I
Cost: preclinical
Cycle time: lead optimization
Cost: target-to-hit
Cycle time: preclinical
p(TS): hit-to-lead
Cost: submission to launch
Cycle time: hit-to-lead
p(TS): target-to-hit
Cycle time: target-to-hit
34%
70%
$10 million
2.5 years
54%
91%
2.5 years
$40 million
$150 million
1.5 years
$15 million
69%
$2.5 million
85%
1.5 years
$5 million
2 years
$1 million
1 year
75%
$40 million
1.5 years
80%
1 year
Capitalized cost per launch (US$ millions)
$1,200 $1,400 $1,600 $1,800 $2,000 $2,200 $2,400
Parameter Baseline value
25%
60%
$15
3.75
45%
80%
3.75
$60
$225
2.25
$22.5
60%
$3.75
75%
2.25
$7.5
3.0
$1.5
1.5
65%
$60
2.25
70%
1.5
50%
80%
$5
1.25
65%
100%
1.25
$20
$75
0.75
$7.5
80%
$1.25
95%
0.75
$2.5
1.0
$0.5
0.5
85%
$20
0.75
90%
0.5
determinant of overall R&D efficiency. In our baseline
model, Phase II p(TS) is 34% (that is, 66% of compounds
entering Phase II fail prior to Phase III). If Phase II attri-
tion increases to 75% (a p(TS) of only 25%), then the
cost per NME increases to $2.3 billion, or an increase of
29%. conversely, if Phase II attrition decreases from 66%
to 50% (that is, a p(TS) of 50%), then the cost per NME
decreases by 25% to $1.33 billion. Similarly, our baseline
value of p(TS) for Phase III molecules is 70%; that is,
an attrition rate of 30%. If Phase III attrition increases
to 40%, then the cost per NME will increase by 16% to
$2.07 billion. conversely, if Phase III attrition can be
reduced to 20% (80% p(TS)), then the cost per NME
will be reduced by 12% to $1.56 billion (FIG. 3).
combining the impact of these increases or decreases
in Phase II and Phase III attrition illustrates the profound
effect of late-stage attrition on R&D efficiency. At the
higher end of the Phase II and III attrition rates discussed
above, the cost of an NME increases from our baseline
case by almost $0.9 billion to $2.7 billion, whereas at the
lower end of the above attrition rates for Phase II and III,
the cost per NME is reduced to $1.17 billion.
It is clear from our analyses that improving R&D effi-
ciency and productivity will depend strongly on reducing
Phase II and III attrition. Unfortunately, industry trends
suggest that both Phase II and III attrition are increas-
ing
9,1921
, given both the more unprecedented nature of
the drug targets being pursued, as well as heightened
scrutiny and concerns about drug safety and the necessity
of demonstrating a highly desirable benefit-to-risk ratio
and health outcome for new medicines. however, main-
taining sufficient WIP while simultaneously reducing CT
and C will also be necessary to improve R&D efficiency.
we discuss these aspects first, before considering strategies
to reduce attrition in depth.
Work in process (WIP). we have already emphasized
the importance of having sufficient WIP at each phase
of drug discovery and development, and have suggested
that insufficient WIP, especially in discovery and the
early phases of clinical development has contributed
to the decline in NME approvals. To further illustrate
this point and again demonstrate the impact of Phase II
and Phase III attrition on Phase I WIP requirements, we
have carried out another sensitivity analysis using these
three parameters alone. FIG. 4 shows the impact of varying
Phase II and III attrition on the number of Phase I entries
per year required to launch a single NME annually. If the
p(TS) in Phase II and Phase III are 25% and 50% respec-
tively, approximately 16 compounds must enter Phase I
Figure 3 | r&D productivity model: parametric sensitivity analysis. This parametric sensitivity analysis is created
from an R&D model that calculates the capitalized cost per launch based on assumptions for the models parameters
(the probability of technical success (p(Ts)), cost and cycle time, all by phase). When baseline values for each of the
parameters are applied, the model calculates a capitalized cost per launch of Us$1,778 million (see supplementary
information s2 (box) for details). This forms the spine of the sensitivity analysis (tornado diagram). The analysis varies each
of the parameters individually to a high and a low value (while holding all other parameters constant at their base value)
and calculates a capitalized cost per launch based on those new values for that varied parameter. In this analysis, the
values of the parameters are varied from 50% lower and 50% higher relative to the baseline value for cost and cycle time
and approximately plus or minus 10 percentage points for p(Ts). Once cost per launch is calculated for the high and low
values of each parameter, the parameters are ordered from highest to lowest based on the relative magnitude of impact
on the overall cost per launch, and the swings in cost per launch are plotted on the graph. At the top of the graph are the
parameters that have the greatest effect on the cost per launch, with positive effect in blue (for example, reducing cost)
and negative effect in red. Parameters shown lower on the graph have a smaller effect on cost per launch.
nrd_3078_mar10.indd 207 15/2/10 11:08:17
AnAlysi s
P
h
a
s
e

I

e
n
t
r
i
e
s

(
F
H
D
s
)
20
18
16
14
12
10
8
2
4
6
0
15 20 25 30 35 40 45 50 55 60 65
Phase II p(TS) (%)
25%
50%
Phase III
p(TS) (%)
50%
60%
70%
80%
90%
Estimate used
in the model
trials each year to yield 1 NME launch. This number
decreases proportionately as the rates of Phase II and
III attrition decline (FIG. 4). Increasing Phase II and III
p(TS) to 50% and 80%, respectively, reduces the number
of Phase I entries needed per year by almost two thirds.
In other words, and everything else being equal, the
same overall R&D investment should yield 23 times
the number of NMEs at these lower attrition rates.
however, just increasing WIP alone without having
sufficient development capacity will probably have
a highly deleterious effect on CT. littles law
22
, which
relates the throughput or flow of a given project to WIP
and CT, posits that too much WIP will result in increased
CT, especially if development resources become rate-
limiting. The ability and need to accurately estimate the
development resources required (capacity management)
and to carefully and optimally balance WIP throughout
the phases of drug discovery and development must be
emphasized (see Bunch & Schacht
23
for a discussion on
capacity management). In the absence of a substantial
reduction in attrition, pharmaceutical companies must
find more affordable means to increase early-stage WIP
and to expeditiously advance these drug candidates
through the development stages.
how can pharmaceutical companies substantially
increase their pipeline WIP without dramatically
increasing C? First, there must be sufficient WIP in the
early stages of drug discovery, and especially early drug
development (as outlined above). Funding these early-
stage (Phase I and II) compounds, especially in the num-
bers indicated by our model, must come in part from
reducing investments in late-stage development, ideally
by redirecting resources from molecules destined to fail
in Phase III (or even Phase Iv). Given the C and CT of
a single Phase III unit of WIP ($150 million), almost 10
Phase I molecules ($15 million) can be developed for the
same cost, ideally through to proof-of-concept (POc;
see discussion of p(TS) below). Reducing late-phase
attrition through early POc studies (ideally in Phase I)
is therefore crucial to implement this partial solution.
The resources (C) saved by lowering Phase III attrition,
however, must be redirected to fund sufficient discovery
and Phase I/II WIP. Most importantly, advancement into
Phase III should be pursued only for those compounds
with established efficacy (ideally POc in Phase I and
confirmed in Phase II) and a well-defined margin of
safety. Ideally, attrition in Phase III should be due pri-
marily to the emergence of relatively rare and unforeseen
adverse events. Thus, the key is to have sufficient WIP
in the early phases of clinical development to effectively
triage and select molecules that will have a higher p(TS)
in late-stage development.
The question of how to affordably increase WIP,
p(TS) and V without substantially increasing C or
increasing CT due to capacity constraints and lack of
focus is in our view paramount to improve R&D pro-
ductivity. This could be accomplished by transforming
the R&D enterprise from one that is predominantly
owned, operated and fully controlled by a given com-
pany (Fully Integrated Pharmaceutical company or
FIPco) to one that is highly networked, partnered and
leveraged (Fully Integrated Pharmaceutical Network or
FIPNet). Traditionally, large pharmaceutical companies
have pursued the discovery, development, manufacture
and commercialization of their medicines largely by
owning and controlling each component. In part, past
reliance on the FIPco model was as much a necessity
as a choice. Today, however, the opportunity to partner
virtually all elements of R&D through a coordinated and
global network or FIPNet could (if effectively managed)
substantially improve R&D productivity by affordably
enhancing the pipeline from early discovery through to
launch. A FIPNet will theoretically allow greater access
to intellectual property, molecules, capabilities, capital,
knowledge and, of course, talent
2426
. Thus, operated as
a FIPNet, a given R&D organization will be able to play
bigger than its size and better leverage its resources to
increase WIP across the pipeline, as long as the costs
of managing the network do not become prohibitive.
Although a full discussion of such a R&D FIPNet is
beyond the scope of this article, we would emphasize
that it includes many types of partnerships ranging from
function-based outsourcing (for example, toxicology or
clinical development), to lower C and reduce CT in many
cases, to molecule-based risk-sharing partnerships and
even partial ownership or equity investments in smaller
companies or joint ventures with larger companies.
A successfully operated R&D FIPNet should aim to bet-
ter leverage and proportionally decrease C to affordably
increase pipeline WIP, while simultaneously mitigating
financial and technical risks.
Much of the discussion on WIP so far has focused on
ensuring that there is sufficient WIP to increase NME
approvals while addressing the inter-relatedness of the
Figure 4 | effect of Phase ii and iii probability of technical success on the number
of Phase i entries required for one successful launch of a new molecular entity.
This analysis shows the number of Phase I entries (first human dose; FHDs) annually
required to achieve one new molecular entity (nME) launch per year as a result of
modelling baseline assumptions of the probability of technical success (p(Ts)) for the
stages of Phase I and submission-to-launch (54% and 91% respectively) over a range of
p(Ts) for Phase II and Phase III. Each curve represents a different assumption for the
Phase III p(Ts) over the range of 50% to 90%, and the x axis represents varying p(Ts) for
Phase II. The number of Phase I entries (FHDs) annually needed to produce one nME
launch per year can be viewed on the y axis for any combination of Phase III (individual
curve) and Phase II (x axis) p(Ts). For example, at a 70% Phase III p(Ts) (black curve) and
a 35% Phase II p(Ts) (on x axis), the required number of Phase I entries is about 8.5.
nrd_3078_mar10.indd 208 15/2/10 11:08:18
AnAlysi s
p(TS): Phase II
p(TS): Phase III
Cost: lead optimization
Cycle time: Phase III
p(TS): Phase I
p(TS): submission to launch
Cycle time: Phase II
Cost: Phase II
Cost: Phase III
Cycle time: submission to launch
Cost: Phase I
p(TS): preclinical
Cost: hit-to-lead
p(TS): lead optimization
Cycle time: Phase I
Cost: preclinical
Cycle time: lead optimization
Cost: target-to-hit
Cycle time: preclinical
p(TS): hit-to-lead
Cost: submission to launch
Cycle time: hit-to-lead
p(TS): target-to-hit
Cycle time: target-to-hit
34%
70%
$10 million
2.5 years
54%
91%
2.5 years
$40 million
$150 million
1.5 years
$15 million
69%
$2.5 million
85%
1.5 years
$5 million
2 years
$1 million
1 year
75%
$40 million
1.5 years
80%
1 year
Capitalized cost per launch (US$ millions)
$1,200 $1,400 $1,600 $1,800 $2,000 $2,200 $2,400
Parameter Baseline value
25%
60%
$15
3.75
45%
80%
3.75
$60
$225
2.25
$22.5
60%
$3.75
75%
2.25
$7.5
3.0
$1.5
1.5
65%
$60
2.25
70%
1.5
50%
80%
$5
1.25
65%
100%
1.25
$20
$75
0.75
$7.5
80%
$1.25
95%
0.75
$2.5
1.0
$0.5
0.5
85%
$20
0.75
90%
0.5
determinant of overall R&D efficiency. In our baseline
model, Phase II p(TS) is 34% (that is, 66% of compounds
entering Phase II fail prior to Phase III). If Phase II attri-
tion increases to 75% (a p(TS) of only 25%), then the
cost per NME increases to $2.3 billion, or an increase of
29%. conversely, if Phase II attrition decreases from 66%
to 50% (that is, a p(TS) of 50%), then the cost per NME
decreases by 25% to $1.33 billion. Similarly, our baseline
value of p(TS) for Phase III molecules is 70%; that is,
an attrition rate of 30%. If Phase III attrition increases
to 40%, then the cost per NME will increase by 16% to
$2.07 billion. conversely, if Phase III attrition can be
reduced to 20% (80% p(TS)), then the cost per NME
will be reduced by 12% to $1.56 billion (FIG. 3).
combining the impact of these increases or decreases
in Phase II and Phase III attrition illustrates the profound
effect of late-stage attrition on R&D efficiency. At the
higher end of the Phase II and III attrition rates discussed
above, the cost of an NME increases from our baseline
case by almost $0.9 billion to $2.7 billion, whereas at the
lower end of the above attrition rates for Phase II and III,
the cost per NME is reduced to $1.17 billion.
It is clear from our analyses that improving R&D effi-
ciency and productivity will depend strongly on reducing
Phase II and III attrition. Unfortunately, industry trends
suggest that both Phase II and III attrition are increas-
ing
9,1921
, given both the more unprecedented nature of
the drug targets being pursued, as well as heightened
scrutiny and concerns about drug safety and the necessity
of demonstrating a highly desirable benefit-to-risk ratio
and health outcome for new medicines. however, main-
taining sufficient WIP while simultaneously reducing CT
and C will also be necessary to improve R&D efficiency.
we discuss these aspects first, before considering strategies
to reduce attrition in depth.
Work in process (WIP). we have already emphasized
the importance of having sufficient WIP at each phase
of drug discovery and development, and have suggested
that insufficient WIP, especially in discovery and the
early phases of clinical development has contributed
to the decline in NME approvals. To further illustrate
this point and again demonstrate the impact of Phase II
and Phase III attrition on Phase I WIP requirements, we
have carried out another sensitivity analysis using these
three parameters alone. FIG. 4 shows the impact of varying
Phase II and III attrition on the number of Phase I entries
per year required to launch a single NME annually. If the
p(TS) in Phase II and Phase III are 25% and 50% respec-
tively, approximately 16 compounds must enter Phase I
Figure 3 | r&D productivity model: parametric sensitivity analysis. This parametric sensitivity analysis is created
from an R&D model that calculates the capitalized cost per launch based on assumptions for the models parameters
(the probability of technical success (p(Ts)), cost and cycle time, all by phase). When baseline values for each of the
parameters are applied, the model calculates a capitalized cost per launch of Us$1,778 million (see supplementary
information s2 (box) for details). This forms the spine of the sensitivity analysis (tornado diagram). The analysis varies each
of the parameters individually to a high and a low value (while holding all other parameters constant at their base value)
and calculates a capitalized cost per launch based on those new values for that varied parameter. In this analysis, the
values of the parameters are varied from 50% lower and 50% higher relative to the baseline value for cost and cycle time
and approximately plus or minus 10 percentage points for p(Ts). Once cost per launch is calculated for the high and low
values of each parameter, the parameters are ordered from highest to lowest based on the relative magnitude of impact
on the overall cost per launch, and the swings in cost per launch are plotted on the graph. At the top of the graph are the
parameters that have the greatest effect on the cost per launch, with positive effect in blue (for example, reducing cost)
and negative effect in red. Parameters shown lower on the graph have a smaller effect on cost per launch.
nrd_3078_mar10.indd 207 15/2/10 11:08:17
AnAlysi s
new drug may require many months or years of continued
treatment. Moreover, even for drugs for which efficacy can
be established relatively quickly (for example, in diabetes),
establishing safety, especially cardiovascular safety, may
require many months or years of treatment. Thus, con-
sidering disease-specific CTs when selecting molecules or
projects for clinical development is crucial for determin-
ing the overall movement or flow of a given pipeline.
The second approach to reducing CT is the iden-
tification and optimization of the critical chain of
project tasks. This approach is based on the concepts
put forth in the Theory of constraints by Goldratt
31
,
wherein the critical chain of tasks is identified and then
the sequence, priority and extent of parallel execution
is determined prior to project or clinical trial execu-
tion. This approach to project management carefully
monitors the critical chain and as delays are encoun-
tered and incurred, the chain can be reconfigured (in
real time) with the aim of recovering any overall project
delay. In our experience, when this approach is lever-
aged against an entire R&D portfolio, it can result in
markedly improved CTs. Thus, CT reduction for the
unit work processes that allow for the completion of
critical chain tasks is essential.
Once a critical chain of tasks is established, there is
the opportunity to improve the business and research
processes used to deliver the tasks within the project
plan. Each task relies on a process, which is often repeat-
able and therefore suitable for optimization. Process
improvement techniques such as Six Sigma can often
be applied to achieve process CT reduction. Examples
of this include the introduction of automation for the
rapid processing of samples, the use of information tech-
nology to support real-time data capture or optimiza-
tion of clinical trial enrolment. however, elimination
of non-value-added activities from the critical chain of
tasks is also essential. Examination of the R&D pro cess
can result in the identification of activities and tasks that
add little or minimal value to desired outputs or out-
comes. Often, these tasks are left over from historical
approaches to R&D that have become obsolete, or are
inefficient bureaucratic procedures associated with the
management of R&D that have become institutional-
ized in an organization, such as multiple serial project
reviews by the leadership of each R&D function. when
such non-value added tasks reside on the critical chain
of the project, they will delay the project from achieving
its milestones.
Finally, the use of adaptive and seamless Phase II and
III study designs have proven extremely useful in reduc-
ing clinical development CT, generally by reducing or
even eliminating non-value-added wait times between
phases of development
32
. Selecting those molecules and
disease indications for which adaptive and seamless
late-stage clinical development is possible can markedly
decrease CT and improve P.
As can be seen from our analysis (FIG. 3), not all CT
reductions are equal in their ability to reduce overall cost
per NME. when applying the approaches above, effort
should be prioritized according to both the impact and
their feasibility of achieving the proposed CT reduction.
In addition to reducing the overall cost per NME, CT
reductions have a positive effect on V associated with a
molecule by allowing greater time on the market under
a protected patent.
Cost (C). like CT, an understanding of the C of R&D
activities can lead to interventions that can reduce the
overall operating expenses required to deliver a success-
ful molecule to the market. The C of an R&D programme
can be split into three categories: direct spend on value-
added tasks, direct spend on non-value-added tasks and
overheads. Focusing on the cost of individual activities
that achieve the objectives of an R&D project can yield
important opportunities to reduce R&D expense. As
R&D involves both the production of knowledge and
materials, one must consider how the expense associated
with an individual task achieves both production of the
requisite information and of the materials that are crucial
for downstream activities. leveraging new technology
(such as software tools or laboratory automation) can
lead to reduced costs, but care must be taken to eliminate
the more tedious and costly methods while introducing
new approaches, otherwise costs can become cumula-
tive. As in the case of CT reduction, process improve-
ment methods such as Six Sigma can yield cost savings
as well. Finally, accessing lower-cost sources of labour
(through outsourcing for example) can yield savings,
assuming no compromise of CT or quality.
Identifying and eliminating non-value-added tasks,
especially non-molecule C, from R&D programmes can
also yield cost-saving opportunities just as it can improve
CT. Furthermore, any sufficiently large R&D organiza-
tion will carry the costs associated with employing and
coordinating its employee base. As these overhead costs
are spread over the entire portfolio of R&D projects, they
have an incremental but often consequential effect on the
C of R&D. Because many of the activities that are clas-
sified as overheads are indeed essential to the success-
ful operation of a business, they cannot be eliminated
entirely. however, in our experience, these activities can
be rationalized and reduced to carry the least amount of
C considered necessary. Such overhead costs are typically
more prevalent in larger, more mature organizations.
Unit cost reductions, like CT reductions, can be lever-
aged to improve productivity (FIG. 3). however, when cut-
ting costs, it is important to be aware not to cut corners
that might reduce the V or p(TS) of the portfolio.
Probability of technical success (p(TS)). There is little
doubt that reducing the attrition rate of drug candidates
in clinical development represents the greatest challenge
and opportunity for pharmaceutical R&D, and arguably
for sustaining the viability of the entire industry. As our
sensitivity analyses show (FIG. 3), reducing Phase II and
III attrition are the strongest levers for improving R&D
efficiency and reducing the costs per NME.
Any given molecule fails because of either technical or
non-technical reasons. Non-technical attrition occurs for
various reasons, such as a change of strategy or termina-
tion of research in a therapeutic area for scientific or com-
mercial reasons, and will not be discussed further, except
nrd_3078_mar10.indd 210 15/2/10 11:08:18
AnAlysi s
Imatinib and trastuzumab
Imatinib blocks the activity
of BCRABL, a deregulated
tyrosine kinase that results
from a chromosomal
translocation in patients with
chronic myelogenous
leukaemia, and trastuzumab
blocks the activity of HER2/
neu, a receptor tyrosine kinase
that is often overexpressed in
patients with breast cancer.
Patients that are most likely
to benefit from each drug can
be readily identified before
initiating treatment on the
basis of the associated
biomarkers, which has been
invaluable in the development
of both drugs and in guiding
their use.
Six Sigma
A quality management tool
that is used to improve the
quality of manufacturing and
business processes by first
identifying and removing the
causes of errors or defects,
as well as by minimizing
variability.
key productivity parameters. however, all R&D WIP
represents expense, albeit an expense that is necessary
to generate innovative medicines. Publicly traded com-
panies have a limit to how much they can invest in R&D
as a percent of sales and thus the volume of WIP that is
affordable. A common mistake is to focus on increasing
WIP without regarding V and p(TS). This can easily
happen when metrics and reward systems focus solely
on the amount of WIP. Ultimately, more WIP without
increases in the V or p(TS) associated with the WIP will
ensure more output, but increases in R&D expenses will
grow proportionately so the strategy will not result in
greater productivity. If V and/or p(TS) could be increased,
more output or value would actually be derived from less
WIP (and expense). At Eli lilly, we have chosen to focus
our view of WIP by adjusting our models to include
dimensions of V and p(TS), ensuring we are measuring
the value of WIP, not just the amount.
Value (V). The effectiveness of drug discovery and
development must be increasingly quantified by estab-
lishing at the time of product approval a highly desirable
health outcome or economic benefit that can be objec-
tively measured in various ways (for example, decreased
mortality, morbidity and reduced hospitalizations).
Thus, to increase R&D effectiveness it is important to
fully understand the ultimate value of a project very early
in development and know how this information can be
leveraged in individual clinical plans and trade-offs in
portfolio decision making.
The determinants of overall value are likely to be
different depending on the perspective represented.
Patients, caregivers, treating physicians and payers may
apply different criteria to determine the value of new
drug therapies. For instance, payers are increasingly
interested in data from clinical trials with active compa-
rators to establish the benefit of a new medicine versus
other therapies (especially generic drugs), whereas regu-
lators primarily depend on placebo-controlled studies to
establish efficacy and safety. To optimize the value of a
portfolio, it is important that optimal development plans
and strategies for each project are formulated early and
well before pivotal registration trials are initiated.
we assume that if a given drug treatment provides a
higher benefit-to-risk ratio, the potential value of that
treatment will be higher
27
. Maximizing patient benefit-
to-risk ratio and thus potential value can be challenging
in many diseases due to substantial clinical and biological
heterogeneity. To increase the benefit (and reduce risks)
of a drug treatment, it is often important to aim to per-
sonalize or tailor the use of the therapy
28
. An important
aid in the pursuit of tailored therapeutics is the identifi-
cation of biomarkers that can be used to diagnose the dis-
ease and/or identify treatment responders, ideally before
or at least after the drug has been given. Biomarkers can
also be used to select the right patients, right dose
and right duration of treatment, or to avoid exposing
patients at risk of a serious adverse event. The use of
biomarkers to appropriately select patient subpopula-
tions will also have a positive effect on p(TS) in the later
stages of clinical development.
To optimize biomarker development and capture
value, it is imperative to ensure biomarkers are devel-
oped early and ideally are commercially available at the
time of launch. Tailoring therapies to specific patient
populations that are predicted to respond on the basis
of the presence of a biomarker can be used to reduce
development costs C, as illustrated by the development
of imatinib (Gleevec; Novartis) or in stratifying patient
populations for both clinical development and commer-
cialization, as for trastuzumab (herceptin; Genentech/
Roche). Although many industry pundits have opined
that such market segmentation will reduce commercial
returns for a given medicine (by reducing market share),
so far, the increase in benefit-to-risk (and thus V) has
more than offset the reduced market share for many new
targeted medicines, especially in oncology.
Cycle time (CT). As is evident from our sensitivity
analysis (FIG. 3), reducing both Phase III and Phase II
CT are also important levers for improving R&D effi-
ciency. Reducing either Phase III or Phase II CT by 50%
from our baseline value of 2.5 years to 1.25 years would
reduce the C per NME by about $200 million. Similar
results were reported by DiMasi
29
. Although we think
such reductions in Phase II and III CT are unrealistic,
even more modest reductions in CT will nonetheless
have a significant impact on R&D efficiency. Finally,
some aspects of R&D CT, such as the time required for
regulatory review (submission to launch in FIG. 3) are less
amenable to intervention.
Reductions in R&D CT can be achieved in several
ways and have long been a goal in the management of any
production system. In predictable production systems,
such as product manufacturing, CT reduction principles
have been mastered and are routinely applied through
techniques such as Six Sigma
30
. In less predictable sys-
tems, such as pharmaceutical R&D, such approaches to
CT reduction cannot be as broadly adapted, but many of
the same principles can be readily applied
30
.
considerations for reducing the CT of each phase
of discovery and development are often project- and
phase-specific. Each phase of development consists
of a collection of unit processes, or individual tasks
that consume time and resources. The arrangement of
these tasks in time and sequence can often be unique
to the individual research project owing to the signifi-
cant variability introduced by the state of the science
associated with the R&D project in question. Because
of this inherent variability, CT reduction at a macro
level is best achieved through task- and project-specific
interventions aimed at reducing non-value added tasks
and wait times associated with completing the value-
added tasks.
we have identified four key approaches for reducing
project-specific CTs. The first is portfolio selection.
when selecting a portfolio of R&D projects, the inclu-
sion of overall project CT as an element of an integrated
strategy of portfolio selection should aid in reducing over-
all CTs. For example, the CT of pivotal clinical trials are
often a function of the disease state or indication being
pursued. In some cases, demonstrating the efficacy of a
nrd_3078_mar10.indd 209 15/2/10 11:08:18
AnAlysi s
new drug may require many months or years of continued
treatment. Moreover, even for drugs for which efficacy can
be established relatively quickly (for example, in diabetes),
establishing safety, especially cardiovascular safety, may
require many months or years of treatment. Thus, con-
sidering disease-specific CTs when selecting molecules or
projects for clinical development is crucial for determin-
ing the overall movement or flow of a given pipeline.
The second approach to reducing CT is the iden-
tification and optimization of the critical chain of
project tasks. This approach is based on the concepts
put forth in the Theory of constraints by Goldratt
31
,
wherein the critical chain of tasks is identified and then
the sequence, priority and extent of parallel execution
is determined prior to project or clinical trial execu-
tion. This approach to project management carefully
monitors the critical chain and as delays are encoun-
tered and incurred, the chain can be reconfigured (in
real time) with the aim of recovering any overall project
delay. In our experience, when this approach is lever-
aged against an entire R&D portfolio, it can result in
markedly improved CTs. Thus, CT reduction for the
unit work processes that allow for the completion of
critical chain tasks is essential.
Once a critical chain of tasks is established, there is
the opportunity to improve the business and research
processes used to deliver the tasks within the project
plan. Each task relies on a process, which is often repeat-
able and therefore suitable for optimization. Process
improvement techniques such as Six Sigma can often
be applied to achieve process CT reduction. Examples
of this include the introduction of automation for the
rapid processing of samples, the use of information tech-
nology to support real-time data capture or optimiza-
tion of clinical trial enrolment. however, elimination
of non-value-added activities from the critical chain of
tasks is also essential. Examination of the R&D pro cess
can result in the identification of activities and tasks that
add little or minimal value to desired outputs or out-
comes. Often, these tasks are left over from historical
approaches to R&D that have become obsolete, or are
inefficient bureaucratic procedures associated with the
management of R&D that have become institutional-
ized in an organization, such as multiple serial project
reviews by the leadership of each R&D function. when
such non-value added tasks reside on the critical chain
of the project, they will delay the project from achieving
its milestones.
Finally, the use of adaptive and seamless Phase II and
III study designs have proven extremely useful in reduc-
ing clinical development CT, generally by reducing or
even eliminating non-value-added wait times between
phases of development
32
. Selecting those molecules and
disease indications for which adaptive and seamless
late-stage clinical development is possible can markedly
decrease CT and improve P.
As can be seen from our analysis (FIG. 3), not all CT
reductions are equal in their ability to reduce overall cost
per NME. when applying the approaches above, effort
should be prioritized according to both the impact and
their feasibility of achieving the proposed CT reduction.
In addition to reducing the overall cost per NME, CT
reductions have a positive effect on V associated with a
molecule by allowing greater time on the market under
a protected patent.
Cost (C). like CT, an understanding of the C of R&D
activities can lead to interventions that can reduce the
overall operating expenses required to deliver a success-
ful molecule to the market. The C of an R&D programme
can be split into three categories: direct spend on value-
added tasks, direct spend on non-value-added tasks and
overheads. Focusing on the cost of individual activities
that achieve the objectives of an R&D project can yield
important opportunities to reduce R&D expense. As
R&D involves both the production of knowledge and
materials, one must consider how the expense associated
with an individual task achieves both production of the
requisite information and of the materials that are crucial
for downstream activities. leveraging new technology
(such as software tools or laboratory automation) can
lead to reduced costs, but care must be taken to eliminate
the more tedious and costly methods while introducing
new approaches, otherwise costs can become cumula-
tive. As in the case of CT reduction, process improve-
ment methods such as Six Sigma can yield cost savings
as well. Finally, accessing lower-cost sources of labour
(through outsourcing for example) can yield savings,
assuming no compromise of CT or quality.
Identifying and eliminating non-value-added tasks,
especially non-molecule C, from R&D programmes can
also yield cost-saving opportunities just as it can improve
CT. Furthermore, any sufficiently large R&D organiza-
tion will carry the costs associated with employing and
coordinating its employee base. As these overhead costs
are spread over the entire portfolio of R&D projects, they
have an incremental but often consequential effect on the
C of R&D. Because many of the activities that are clas-
sified as overheads are indeed essential to the success-
ful operation of a business, they cannot be eliminated
entirely. however, in our experience, these activities can
be rationalized and reduced to carry the least amount of
C considered necessary. Such overhead costs are typically
more prevalent in larger, more mature organizations.
Unit cost reductions, like CT reductions, can be lever-
aged to improve productivity (FIG. 3). however, when cut-
ting costs, it is important to be aware not to cut corners
that might reduce the V or p(TS) of the portfolio.
Probability of technical success (p(TS)). There is little
doubt that reducing the attrition rate of drug candidates
in clinical development represents the greatest challenge
and opportunity for pharmaceutical R&D, and arguably
for sustaining the viability of the entire industry. As our
sensitivity analyses show (FIG. 3), reducing Phase II and
III attrition are the strongest levers for improving R&D
efficiency and reducing the costs per NME.
Any given molecule fails because of either technical or
non-technical reasons. Non-technical attrition occurs for
various reasons, such as a change of strategy or termina-
tion of research in a therapeutic area for scientific or com-
mercial reasons, and will not be discussed further, except
nrd_3078_mar10.indd 210 15/2/10 11:08:18
AnAlysi s
Imatinib and trastuzumab
Imatinib blocks the activity
of BCRABL, a deregulated
tyrosine kinase that results
from a chromosomal
translocation in patients with
chronic myelogenous
leukaemia, and trastuzumab
blocks the activity of HER2/
neu, a receptor tyrosine kinase
that is often overexpressed in
patients with breast cancer.
Patients that are most likely
to benefit from each drug can
be readily identified before
initiating treatment on the
basis of the associated
biomarkers, which has been
invaluable in the development
of both drugs and in guiding
their use.
Six Sigma
A quality management tool
that is used to improve the
quality of manufacturing and
business processes by first
identifying and removing the
causes of errors or defects,
as well as by minimizing
variability.
key productivity parameters. however, all R&D WIP
represents expense, albeit an expense that is necessary
to generate innovative medicines. Publicly traded com-
panies have a limit to how much they can invest in R&D
as a percent of sales and thus the volume of WIP that is
affordable. A common mistake is to focus on increasing
WIP without regarding V and p(TS). This can easily
happen when metrics and reward systems focus solely
on the amount of WIP. Ultimately, more WIP without
increases in the V or p(TS) associated with the WIP will
ensure more output, but increases in R&D expenses will
grow proportionately so the strategy will not result in
greater productivity. If V and/or p(TS) could be increased,
more output or value would actually be derived from less
WIP (and expense). At Eli lilly, we have chosen to focus
our view of WIP by adjusting our models to include
dimensions of V and p(TS), ensuring we are measuring
the value of WIP, not just the amount.
Value (V). The effectiveness of drug discovery and
development must be increasingly quantified by estab-
lishing at the time of product approval a highly desirable
health outcome or economic benefit that can be objec-
tively measured in various ways (for example, decreased
mortality, morbidity and reduced hospitalizations).
Thus, to increase R&D effectiveness it is important to
fully understand the ultimate value of a project very early
in development and know how this information can be
leveraged in individual clinical plans and trade-offs in
portfolio decision making.
The determinants of overall value are likely to be
different depending on the perspective represented.
Patients, caregivers, treating physicians and payers may
apply different criteria to determine the value of new
drug therapies. For instance, payers are increasingly
interested in data from clinical trials with active compa-
rators to establish the benefit of a new medicine versus
other therapies (especially generic drugs), whereas regu-
lators primarily depend on placebo-controlled studies to
establish efficacy and safety. To optimize the value of a
portfolio, it is important that optimal development plans
and strategies for each project are formulated early and
well before pivotal registration trials are initiated.
we assume that if a given drug treatment provides a
higher benefit-to-risk ratio, the potential value of that
treatment will be higher
27
. Maximizing patient benefit-
to-risk ratio and thus potential value can be challenging
in many diseases due to substantial clinical and biological
heterogeneity. To increase the benefit (and reduce risks)
of a drug treatment, it is often important to aim to per-
sonalize or tailor the use of the therapy
28
. An important
aid in the pursuit of tailored therapeutics is the identifi-
cation of biomarkers that can be used to diagnose the dis-
ease and/or identify treatment responders, ideally before
or at least after the drug has been given. Biomarkers can
also be used to select the right patients, right dose
and right duration of treatment, or to avoid exposing
patients at risk of a serious adverse event. The use of
biomarkers to appropriately select patient subpopula-
tions will also have a positive effect on p(TS) in the later
stages of clinical development.
To optimize biomarker development and capture
value, it is imperative to ensure biomarkers are devel-
oped early and ideally are commercially available at the
time of launch. Tailoring therapies to specific patient
populations that are predicted to respond on the basis
of the presence of a biomarker can be used to reduce
development costs C, as illustrated by the development
of imatinib (Gleevec; Novartis) or in stratifying patient
populations for both clinical development and commer-
cialization, as for trastuzumab (herceptin; Genentech/
Roche). Although many industry pundits have opined
that such market segmentation will reduce commercial
returns for a given medicine (by reducing market share),
so far, the increase in benefit-to-risk (and thus V) has
more than offset the reduced market share for many new
targeted medicines, especially in oncology.
Cycle time (CT). As is evident from our sensitivity
analysis (FIG. 3), reducing both Phase III and Phase II
CT are also important levers for improving R&D effi-
ciency. Reducing either Phase III or Phase II CT by 50%
from our baseline value of 2.5 years to 1.25 years would
reduce the C per NME by about $200 million. Similar
results were reported by DiMasi
29
. Although we think
such reductions in Phase II and III CT are unrealistic,
even more modest reductions in CT will nonetheless
have a significant impact on R&D efficiency. Finally,
some aspects of R&D CT, such as the time required for
regulatory review (submission to launch in FIG. 3) are less
amenable to intervention.
Reductions in R&D CT can be achieved in several
ways and have long been a goal in the management of any
production system. In predictable production systems,
such as product manufacturing, CT reduction principles
have been mastered and are routinely applied through
techniques such as Six Sigma
30
. In less predictable sys-
tems, such as pharmaceutical R&D, such approaches to
CT reduction cannot be as broadly adapted, but many of
the same principles can be readily applied
30
.
considerations for reducing the CT of each phase
of discovery and development are often project- and
phase-specific. Each phase of development consists
of a collection of unit processes, or individual tasks
that consume time and resources. The arrangement of
these tasks in time and sequence can often be unique
to the individual research project owing to the signifi-
cant variability introduced by the state of the science
associated with the R&D project in question. Because
of this inherent variability, CT reduction at a macro
level is best achieved through task- and project-specific
interventions aimed at reducing non-value added tasks
and wait times associated with completing the value-
added tasks.
we have identified four key approaches for reducing
project-specific CTs. The first is portfolio selection.
when selecting a portfolio of R&D projects, the inclu-
sion of overall project CT as an element of an integrated
strategy of portfolio selection should aid in reducing over-
all CTs. For example, the CT of pivotal clinical trials are
often a function of the disease state or indication being
pursued. In some cases, demonstrating the efficacy of a
nrd_3078_mar10.indd 209 15/2/10 11:08:18
AnAlysi s
Preclinical
development
Launch
$
Phase I
$
Phase II
$ $
Phase III
R&D sweet spot
$ $
$ $
CS
FHD
FED
PD
Test each
scarce molecule
thoroughly
Scarcity
of drug
discovery
a Traditional
Preclinical
development
Launch
POC Confirmation,
dose finding
Commercialization
Higher p(TS)
CS
FHD
PD
Increase critical information
content early to shift attrition
to cheaper phase
Use savings from
shifted attrition to re-invest
in the R&D sweet spot
Abundance
of drug
discovery
b Quick win, fast fail
Chorus
A virtual approach to drug
development that is primarily
focused on establishing early
proof-of-concept in humans
(ideally in Phase I) to reduce
attrition at later stages. Chorus
cost estimate quoted in the
text was calculated including
the direct and indirect costs for
21 molecules in the Chorus
portfolio from 20052008
and one from 2004.
increase PcSK9 levels, reducing their potential efficacy
in further lowering lDlcholesterol
39
. Thus, inhibiting
PcSK9 function could dramatically (and safely) reduce
lDlcholesterol as well as augment the beneficial effects
of statins. PcSK9 inhibition can conceivably be accom-
plished in several ways, but perhaps the most direct is
with a neutralizing monoclonal antibody, which as noted
earlier, could be developed relatively rapidly and with
limited off-target toxicity. Such an antibody has recently
been reported by chan and colleagues
40
to dramatically
lower lDlcholesterol in both mice and non-human pri-
mates. It should now be relatively straightforward to create
a humanized (or fully human) neutralizing monoclonal
antibody against PcSK9 for human testing. Importantly,
POc with this antibody can be achieved rapidly in Phase I
studies by simply measuring lDlcholesterol, perhaps as
soon as following administration of a single dose. Recent
clinical and epidemiological studies strongly suggest that
lowering lDlcholesterol below currently accepted tar-
gets (with or without a statin) may further reduce the
incidence of chD
41
. It is quite possible that a PcSK9 anti-
body or some other PcSK9 antagonist could dramatically
(and safely) lower lDlcholesterol and the risk of chD.
however, the key point here is that the validation of this
target by human genetics and preclinical studies using
a potential therapeutic agent in animals, coupled with
the ability to establish POc very early in the clinic, will
undoubtedly increase the p(TS) of this molecule before it
advances into Phase II and III trials.
The second example is the voltage-gated sodium
channel type IX alpha subunit (also known as Na
v
1.7 or
ScN9A). like PcSK9, both gain-of-function and loss-
of-function mutations in Na
v
1.7 have been identified
42
.
These mutations result in either increased sensitivity to
painful stimuli (paroxysmal extreme pain disorder and
primary erythermalgia) or channelopathy-associated
insensitivity to pain, respectively
42
. The latter is a rare
autosomal recessive disorder in which affected indi-
viduals are insensitive to physical pain from birth.
Importantly, these individuals have otherwise normal
sensory perception, although the consequences of com-
plete and persistent insensitivity to pain can be quite
detri mental
42
. Nonetheless, a selective and reversible Na
v

1.7 antagonist, perhaps one that inhibits channel activity
in a state-dependent manner, will probably prove to be
a very effective topical or oral analgesic agent
43
. Non-
selective sodium channel blockers, such as the local
anesthetic lidocaine and certain anticonvulsants such
as phenytoin and carbamezepine, may in part work
through Na
v
1.7, but finding more selective antagonists
could revolutionize the treatment of pain
43
. Moreover,
as was the case for PcSK9, clinical POc can be rapidly
obtained in Phase I using various human pain models.
These two drug targets have one important feature
in common. Mutations in their genes cause a loss of
function that results in a clinical phenotype (low lDl
cholesterol or insensitivity to pain) that could be mim-
icked by a drug with potentially desirable therapeutic
consequences. The validation of these two targets derives
primarily from human genetics, coupled with a very
good (but admittedly still incomplete) understanding of
their biological mechanisms and clinical consequences.
we estimate that there are now at least a dozen or more
drug targets with similar human validation and this
number will undoubtedly grow as advances in functional
genomics continue to materialize.
Using these two approaches (better target validation
and early POc studies) we have estimated that the p(TS)
of Phase II compounds can be increased to approxi-
mately 50%. As can be seen in our sensitivity analysis
(FIG. 3), reducing Phase II attrition by this much will by
itself lower the cost per NME by approximately 30%.
In our view, resolution of technical uncertainty early
in development, especially whether or not a molecule
engages its target and has desired pharmacological activity
in humans is necessary to improve R&D productivity.
we refer to this as our quick win, fast fail paradigm of
drug development (FIG. 5), whereby the sweet spot of
R&D resides prior to Phase II with a heavy emphasis
on having sufficient discovery capacity and capability to
assure selection of validated targets, and sufficient dis-
covery WIP to yield enough good Phase-I-ready mole-
cules. A resolute focus and portfolio metric is to establish
POc in the clinic very early, preferably in Phase I. This
can reliably be achieved for most, but obviously not all,
drug candidates.
we have implemented an alternative clinical devel-
opment model called Chorus
44
, which has served as a
highly cost-effective means for establishing POc and
has, in part, helped to increase the estimated p(TS) of
Figure 5 | The quick win, fast fail drug development paradigm. This figure illustrates
the traditional paradigm of drug development (a) contrasted with an alternative
development paradigm referred to as quick win, fast fail (b). In this alternative, technical
uncertainty is intentionally decreased before the expensive later development stages
(Phase II and Phase III) through the establishment of proof-of-concept (POC). This results
in a reduced number of new molecular entities (nMEs) advancing into Phase II and III,
but those that do advance have a higher probability of success (p(Ts)) and launch. The
savings gained from costly investment in late-stage R&D failures are re-invested in R&D
to further enhance R&D productivity. Cs, candidate selection; FED, first efficacy dose;
FHD, first human dose; PD, product decision.
nrd_3078_mar10.indd 212 15/2/10 11:08:19
AnAlysi s
to say that it can be a considerable (and sometimes sus-
tained) source of C and lowered R&D efficiency. Technical
attrition, on the other hand, occurs when a molecule fails
to meet some important success factor, such as failing to
demonstrate the requisite safety margin in Phase I or II or
expected efficacy in Phase II or III.
compounds fail for many reasons, but some are
more avoidable than others. Poor oral bioavailability,
pharmacokinetic properties or toxicity issues that are
not predicted by animal pharmacology models or by
preclinical ADMET (absorption, distribution, metabo-
lism, excretion and toxicity) studies, resulting in over-
lap of efficacious and toxic doses (and thus lower than
desired margins of safety) are often reasons for Phase I
and Phase II attrition.
Given their highly specific target-binding character-
istics, fully human or humanized monoclonal antibodies
have greatly reduced off-target toxicity compared with
small molecules, which could help reduce attrition. Any
cost-effective approach to predict toxicological liability or
undesirable ADME properties preclinically (especially for
small molecules) will reduce downstream attrition and
several approaches to achieve this are being pursued
33
.
In fact, the number (and percentage) of drug candidates
successfully reaching Phase I has increased recently
18
,
in part due to better preclinical characterization with
improved ADMET properties.
Indeed, in their analysis of attrition rates based on
data from large pharmaceutical companies between
19912000, Kola and landis highlighted that poor phar-
macokinetic properties or bioavailability had become
only a minor cause of overall attrition ( 1020%) by
2000, whereas lack of efficacy and low margins of safety
were the major causes of Phase II and III attrition
19
.
Assuming that approaches for filtering out compounds
with inappropriate drug-like properties have advanced
since the 1990s, it seems likely that the recent increase in
later-stage (especially Phase II) attrition is primarily due
to the unprecedented nature of the drug targets (that is,
the biological mechanisms) being pursued as well as the
increasing safety hurdles (greater benefit-to-risk ratio)
required for approval in most parts of the world.
however, as discussed above, Phase II and Phase III
attrition rates remain unacceptably high: 66% and 30%,
respectively, based on the most recent PBF benchmark-
ing estimates (FIG. 2). Phase II attrition rates in particular
have not improved substantially since those reported for
the 19912000 period (62% attrition rate in Phase II and
45% in Phase III, with some therapeutic areas exhibit-
ing even higher attrition rates)
19
. As highlighted by Kola
and landis, clinical attrition rates during the 1990s were
higher for central nervous system (cNS) disorders and
oncology, with more than 70% of compounds in oncol-
ogy failing in Phase II and 59% failing in Phase III. The
higher failure rates in these areas are in part due to the
relatively unprecedented nature of the drug targets being
pursued and to the lack of animal models with a strong
capacity to predict human efficacy.
There are two key approaches to reduce Phase II
and III attrition, which ideally need to be carried out
in tandem. The first is better target selection (selection
of more validated and druggable targets). The second
is the routine pursuit of early POc studies in the clinic,
especially in Phase I, for which biomarkers and surro-
gate endpoints can often be employed. Although these
two approaches to reducing attrition and increasing pro-
ductivity have been highlighted by others
17,19
, we under-
score their importance, and discuss how they can best
be achieved below.
First, target selection is a key early step in the drug
discovery and development process, as it generally
occurs 1015 years before the launch of a new drug
and initiates the commitment of substantial time and
resources to determine whether the target and the
approach to modulating it are viable. Obviously, target
identification and validation leading up to target selec-
tion are also important factors, and can confer a com-
petitive advantage to an R&D organization. It was widely
predicted that advances in genomics and proteomics,
including those resulting directly from the sequencing of
the human genome, would yield an abundance of drug
targets
34,35
. however, although many new potential drug
targets have been identified by these approaches, far too
few have been fully, or even sufficiently, validated so
far to have had much impact on pharmaceutical R&D
3
.
Nevertheless, we believe that validated targets for drug
discovery are now materializing rapidly.
Target selection is also related to the second key
approach to reducing Phase II and III attrition: estab-
lishing early POc. This requires first choosing a drug
target and disease state for which establishing POc early
(preferably in Phase I) is feasible. The use of biomarkers
or surrogate endpoints (if not clinical endpoints) is often
essential for such POc studies. These biomarkers or sur-
rogates can be as simple as assuring target engagement
(for example, cNS receptor occupancy by positron emis-
sion tomography scanning for neuropsychiatric drugs)
or a desired clinical endpoint (for example, the lowering
of blood glucose for diabetes or low density lipoprotein
(lDl)cholesterol for dyslipidaemia). A number of clini-
cally relevant endpoints can be ascertained in Phase I
(especially in the multidose safety studies or Phase Ib
studies), including analgesia, lipid or glucose lowering
and weight loss, but developing additional biomarkers
of both efficacy and safety for a variety of diseases will
be necessary to make early go/no-go decisions; such
biomarkers are especially needed in oncology.
To illustrate the importance of target selection, cou-
pled with early POc clinical trials, we discuss two recently
validated targets that in our opinion will probably result
in very effective and commercially important drugs. The
first target is the secreted circulating protease proprotein
convertase subtilisin/kexin type 9 (PcSK9). circulating
PcSK9 downregulates hepatic lDl receptors and thus
increases serum lDlcholesterol levels
36
. Gain-of-
function missense mutations in the PcSK9 gene result in
autosomal dominant hypercholesterolaemia and prema-
ture coronary heart disease (chD)
37
. conversely, loss-of-
function mutations are associated with very low serum
lDlcholesterol and a striking reduction in the incidence
of chD
38
. Moreover, hMG coA reductase inhibitors
(statins), which lower lDlcholesterol, simultaneously
nrd_3078_mar10.indd 211 15/2/10 11:08:18
AnAlysi s
Preclinical
development
Launch
$
Phase I
$
Phase II
$ $
Phase III
R&D sweet spot
$ $
$ $
CS
FHD
FED
PD
Test each
scarce molecule
thoroughly
Scarcity
of drug
discovery
a Traditional
Preclinical
development
Launch
POC Confirmation,
dose finding
Commercialization
Higher p(TS)
CS
FHD
PD
Increase critical information
content early to shift attrition
to cheaper phase
Use savings from
shifted attrition to re-invest
in the R&D sweet spot
Abundance
of drug
discovery
b Quick win, fast fail
Chorus
A virtual approach to drug
development that is primarily
focused on establishing early
proof-of-concept in humans
(ideally in Phase I) to reduce
attrition at later stages. Chorus
cost estimate quoted in the
text was calculated including
the direct and indirect costs for
21 molecules in the Chorus
portfolio from 20052008
and one from 2004.
increase PcSK9 levels, reducing their potential efficacy
in further lowering lDlcholesterol
39
. Thus, inhibiting
PcSK9 function could dramatically (and safely) reduce
lDlcholesterol as well as augment the beneficial effects
of statins. PcSK9 inhibition can conceivably be accom-
plished in several ways, but perhaps the most direct is
with a neutralizing monoclonal antibody, which as noted
earlier, could be developed relatively rapidly and with
limited off-target toxicity. Such an antibody has recently
been reported by chan and colleagues
40
to dramatically
lower lDlcholesterol in both mice and non-human pri-
mates. It should now be relatively straightforward to create
a humanized (or fully human) neutralizing monoclonal
antibody against PcSK9 for human testing. Importantly,
POc with this antibody can be achieved rapidly in Phase I
studies by simply measuring lDlcholesterol, perhaps as
soon as following administration of a single dose. Recent
clinical and epidemiological studies strongly suggest that
lowering lDlcholesterol below currently accepted tar-
gets (with or without a statin) may further reduce the
incidence of chD
41
. It is quite possible that a PcSK9 anti-
body or some other PcSK9 antagonist could dramatically
(and safely) lower lDlcholesterol and the risk of chD.
however, the key point here is that the validation of this
target by human genetics and preclinical studies using
a potential therapeutic agent in animals, coupled with
the ability to establish POc very early in the clinic, will
undoubtedly increase the p(TS) of this molecule before it
advances into Phase II and III trials.
The second example is the voltage-gated sodium
channel type IX alpha subunit (also known as Na
v
1.7 or
ScN9A). like PcSK9, both gain-of-function and loss-
of-function mutations in Na
v
1.7 have been identified
42
.
These mutations result in either increased sensitivity to
painful stimuli (paroxysmal extreme pain disorder and
primary erythermalgia) or channelopathy-associated
insensitivity to pain, respectively
42
. The latter is a rare
autosomal recessive disorder in which affected indi-
viduals are insensitive to physical pain from birth.
Importantly, these individuals have otherwise normal
sensory perception, although the consequences of com-
plete and persistent insensitivity to pain can be quite
detri mental
42
. Nonetheless, a selective and reversible Na
v

1.7 antagonist, perhaps one that inhibits channel activity
in a state-dependent manner, will probably prove to be
a very effective topical or oral analgesic agent
43
. Non-
selective sodium channel blockers, such as the local
anesthetic lidocaine and certain anticonvulsants such
as phenytoin and carbamezepine, may in part work
through Na
v
1.7, but finding more selective antagonists
could revolutionize the treatment of pain
43
. Moreover,
as was the case for PcSK9, clinical POc can be rapidly
obtained in Phase I using various human pain models.
These two drug targets have one important feature
in common. Mutations in their genes cause a loss of
function that results in a clinical phenotype (low lDl
cholesterol or insensitivity to pain) that could be mim-
icked by a drug with potentially desirable therapeutic
consequences. The validation of these two targets derives
primarily from human genetics, coupled with a very
good (but admittedly still incomplete) understanding of
their biological mechanisms and clinical consequences.
we estimate that there are now at least a dozen or more
drug targets with similar human validation and this
number will undoubtedly grow as advances in functional
genomics continue to materialize.
Using these two approaches (better target validation
and early POc studies) we have estimated that the p(TS)
of Phase II compounds can be increased to approxi-
mately 50%. As can be seen in our sensitivity analysis
(FIG. 3), reducing Phase II attrition by this much will by
itself lower the cost per NME by approximately 30%.
In our view, resolution of technical uncertainty early
in development, especially whether or not a molecule
engages its target and has desired pharmacological activity
in humans is necessary to improve R&D productivity.
we refer to this as our quick win, fast fail paradigm of
drug development (FIG. 5), whereby the sweet spot of
R&D resides prior to Phase II with a heavy emphasis
on having sufficient discovery capacity and capability to
assure selection of validated targets, and sufficient dis-
covery WIP to yield enough good Phase-I-ready mole-
cules. A resolute focus and portfolio metric is to establish
POc in the clinic very early, preferably in Phase I. This
can reliably be achieved for most, but obviously not all,
drug candidates.
we have implemented an alternative clinical devel-
opment model called Chorus
44
, which has served as a
highly cost-effective means for establishing POc and
has, in part, helped to increase the estimated p(TS) of
Figure 5 | The quick win, fast fail drug development paradigm. This figure illustrates
the traditional paradigm of drug development (a) contrasted with an alternative
development paradigm referred to as quick win, fast fail (b). In this alternative, technical
uncertainty is intentionally decreased before the expensive later development stages
(Phase II and Phase III) through the establishment of proof-of-concept (POC). This results
in a reduced number of new molecular entities (nMEs) advancing into Phase II and III,
but those that do advance have a higher probability of success (p(Ts)) and launch. The
savings gained from costly investment in late-stage R&D failures are re-invested in R&D
to further enhance R&D productivity. Cs, candidate selection; FED, first efficacy dose;
FHD, first human dose; PD, product decision.
nrd_3078_mar10.indd 212 15/2/10 11:08:19
AnAlysi s
to say that it can be a considerable (and sometimes sus-
tained) source of C and lowered R&D efficiency. Technical
attrition, on the other hand, occurs when a molecule fails
to meet some important success factor, such as failing to
demonstrate the requisite safety margin in Phase I or II or
expected efficacy in Phase II or III.
compounds fail for many reasons, but some are
more avoidable than others. Poor oral bioavailability,
pharmacokinetic properties or toxicity issues that are
not predicted by animal pharmacology models or by
preclinical ADMET (absorption, distribution, metabo-
lism, excretion and toxicity) studies, resulting in over-
lap of efficacious and toxic doses (and thus lower than
desired margins of safety) are often reasons for Phase I
and Phase II attrition.
Given their highly specific target-binding character-
istics, fully human or humanized monoclonal antibodies
have greatly reduced off-target toxicity compared with
small molecules, which could help reduce attrition. Any
cost-effective approach to predict toxicological liability or
undesirable ADME properties preclinically (especially for
small molecules) will reduce downstream attrition and
several approaches to achieve this are being pursued
33
.
In fact, the number (and percentage) of drug candidates
successfully reaching Phase I has increased recently
18
,
in part due to better preclinical characterization with
improved ADMET properties.
Indeed, in their analysis of attrition rates based on
data from large pharmaceutical companies between
19912000, Kola and landis highlighted that poor phar-
macokinetic properties or bioavailability had become
only a minor cause of overall attrition ( 1020%) by
2000, whereas lack of efficacy and low margins of safety
were the major causes of Phase II and III attrition
19
.
Assuming that approaches for filtering out compounds
with inappropriate drug-like properties have advanced
since the 1990s, it seems likely that the recent increase in
later-stage (especially Phase II) attrition is primarily due
to the unprecedented nature of the drug targets (that is,
the biological mechanisms) being pursued as well as the
increasing safety hurdles (greater benefit-to-risk ratio)
required for approval in most parts of the world.
however, as discussed above, Phase II and Phase III
attrition rates remain unacceptably high: 66% and 30%,
respectively, based on the most recent PBF benchmark-
ing estimates (FIG. 2). Phase II attrition rates in particular
have not improved substantially since those reported for
the 19912000 period (62% attrition rate in Phase II and
45% in Phase III, with some therapeutic areas exhibit-
ing even higher attrition rates)
19
. As highlighted by Kola
and landis, clinical attrition rates during the 1990s were
higher for central nervous system (cNS) disorders and
oncology, with more than 70% of compounds in oncol-
ogy failing in Phase II and 59% failing in Phase III. The
higher failure rates in these areas are in part due to the
relatively unprecedented nature of the drug targets being
pursued and to the lack of animal models with a strong
capacity to predict human efficacy.
There are two key approaches to reduce Phase II
and III attrition, which ideally need to be carried out
in tandem. The first is better target selection (selection
of more validated and druggable targets). The second
is the routine pursuit of early POc studies in the clinic,
especially in Phase I, for which biomarkers and surro-
gate endpoints can often be employed. Although these
two approaches to reducing attrition and increasing pro-
ductivity have been highlighted by others
17,19
, we under-
score their importance, and discuss how they can best
be achieved below.
First, target selection is a key early step in the drug
discovery and development process, as it generally
occurs 1015 years before the launch of a new drug
and initiates the commitment of substantial time and
resources to determine whether the target and the
approach to modulating it are viable. Obviously, target
identification and validation leading up to target selec-
tion are also important factors, and can confer a com-
petitive advantage to an R&D organization. It was widely
predicted that advances in genomics and proteomics,
including those resulting directly from the sequencing of
the human genome, would yield an abundance of drug
targets
34,35
. however, although many new potential drug
targets have been identified by these approaches, far too
few have been fully, or even sufficiently, validated so
far to have had much impact on pharmaceutical R&D
3
.
Nevertheless, we believe that validated targets for drug
discovery are now materializing rapidly.
Target selection is also related to the second key
approach to reducing Phase II and III attrition: estab-
lishing early POc. This requires first choosing a drug
target and disease state for which establishing POc early
(preferably in Phase I) is feasible. The use of biomarkers
or surrogate endpoints (if not clinical endpoints) is often
essential for such POc studies. These biomarkers or sur-
rogates can be as simple as assuring target engagement
(for example, cNS receptor occupancy by positron emis-
sion tomography scanning for neuropsychiatric drugs)
or a desired clinical endpoint (for example, the lowering
of blood glucose for diabetes or low density lipoprotein
(lDl)cholesterol for dyslipidaemia). A number of clini-
cally relevant endpoints can be ascertained in Phase I
(especially in the multidose safety studies or Phase Ib
studies), including analgesia, lipid or glucose lowering
and weight loss, but developing additional biomarkers
of both efficacy and safety for a variety of diseases will
be necessary to make early go/no-go decisions; such
biomarkers are especially needed in oncology.
To illustrate the importance of target selection, cou-
pled with early POc clinical trials, we discuss two recently
validated targets that in our opinion will probably result
in very effective and commercially important drugs. The
first target is the secreted circulating protease proprotein
convertase subtilisin/kexin type 9 (PcSK9). circulating
PcSK9 downregulates hepatic lDl receptors and thus
increases serum lDlcholesterol levels
36
. Gain-of-
function missense mutations in the PcSK9 gene result in
autosomal dominant hypercholesterolaemia and prema-
ture coronary heart disease (chD)
37
. conversely, loss-of-
function mutations are associated with very low serum
lDlcholesterol and a striking reduction in the incidence
of chD
38
. Moreover, hMG coA reductase inhibitors
(statins), which lower lDlcholesterol, simultaneously
nrd_3078_mar10.indd 211 15/2/10 11:08:18
AnAlysi s
even micro-managing R&D by those with little scientific
medical expertise or experience in R&D may well have
contributed to recent pipeline gaps and reduced R&D
productivity
45
. Moreover, fostering such scientific crea-
tivity and medical opportunism is often challenging, and
sometimes incompatible with the short-term and often
aspirational goals of business-driven enterprises. Most
importantly, in our opinion, there is no substitute for
scientific and clinical intuition for any successful R&D
organization, whether it be in the initial selection of drug
targets, the design of a clinical trial or the interpretation
of clinical data in order to advance a molecule into late-
phase clinical development. Simply stated, good process
will never be a substitute for good people or good science.
however, we firmly believe that both good process and
good science are not only compatible, but together will
yield the greatest return on R&D investments and thus,
have the greatest impact on R&D productivity.
1. OHagan, P. & Farkas, C. Bringing pharma R.&D back
to health. Bain Brief [online], http://www.bain.com/
bainweb/PDFs/cms/Public/BB_Managing_RandD_HC.
pdf (2009).
2. Lindgardt, Z., Reeves, M. & Wallenstein, J. Waking the
giant: business model innovation in the drug industry.
In Vivo 26, 16 (2008).
3. Garnier, J. Rebuilding the R.&D engine in big pharma.
Harvard Bus. Rev. 86, 6876 (2008).
A candid assessment by the former CEO of
GlaxoSmithKline of the downsides of the industrys
process culture.
4. Angell, M. The Truth about Drug Companies: How They
Deceive Us and What to do About It. (Random House
Trade Paperbacks, New York, 2005).
5. Chertkow, J. Trends in pharmaceutical portfolio
management: strategies to maintain profitability
despite adversity. Datamonitor [online] http://www.
datamonitor.com (2008).
6. Murray, A., Berndt, E. R. & Cutler, D. M. Prescription
drug spending trends in the United States: looking
beyond the turning point. Health Affairs w151w160,
(2008).
7. EvaluatePharma Alpha World Preview 2014. Evaluate
Pharma report (2009).
8. Goodman, M. Market watch: Pharma industry
performance metrics: 20072012E. Nature Rev. Drug
Discov. 7, 795 (2008).
9. Mathieu, M. P., ed. Parexels Bio/Pharmaceutical
R&D Statistical Sourcebook 2008/2009.
(Parexel International Corporation, Waltham, 2008).
10. Pisano, G. P. Science Business: the Promise, the
Reality, and the Future of Biotech. (Harvard Business
School Press, Boston, 2006).
An insightful retrospective on the biotech industry,
and the reasons why it has not yet lived up to its
promises.
11. DiMasi, J. A. & Grabowski, H. G. The cost of
biopharmaceutical R&D: Is biotech different?
Manage. Decis. Econ. 28, 469479 (2007).
A contemporary assessment of the costs to
discover, develop and launch new molecular entities.
12. Kola, I. The state of innovation in drug development.
Clin. Pharmacol. Ther. 83, 227230 (2008).
13. Cohen, F. J. Macro trends in pharmaceutical
innovation. Nature Rev. Drug Discov. 4, 7884
(2005).
14. Lichtenberg, F. R. The impact of new drug launches on
longevity: evidence from longitudinal, disease-level
data from 52 countries, 19822001 Int. J. Health
Care Finance Econ. 5, 4743 (2005).
15. Kimball, T. R., McCoy, C. E., Khoury, P. R., Daniels, S. R.
& Dolan, L. M. Obesity, diabetes damage young
arteries, could shorten life. News release, American
Heart Association website [online] http://
americanheart.mediaroom.com/index.
php?s=43&item=740 (2009).
16. KMR Group. Pharmaceutical Benchmarking Forum
[online] http://kmrgroup.com/ForumsPharma.html
(2009).
17. Booth, B. & Zemmil, R. Prospects for productivity.
Nature Rev. Drug Discov. 3, 451456 (2004).
Booth and Zemmil framed the issue of productivity
in the pharmaceutical industry and provide a
qualitative discussion of some of the drivers and
potential solutions.
18. Hu, M., Schultz, K., Sheu, J. & Tschopp, D. The
innovation gap in pharmaceutical drug discovery &
new models for R&D success. [online], http://www.
kellogg.northwestern.edu/academic/biotech/faculty/
articles/NewRDModel.pdf (2007).
19. Kola, I. & Landis, J. Can the pharmaceutical industry
reduce attrition rates? Nature Rev. Drug Discov. 3,
711715 (2004).
This article provides an in-depth analysis of attrition
by phase and therapy area, and offers a scientific
approach to reduce Phase II and Phase III attrition
rates.
20. Innovation or Stagnation? Challenge and Opportunity
on the Critical Path to New Medical Products [online],
http://www.fda.gov/oc/initiatives/criticalpath/
whitepaper.html (2004).
A seminal paper on the hurdles that must be
overcome to reverse the decline in R&D productivity.
21. Peck, R. W. Driving earlier clinical attrition: if you want
to find the needle, burn down the haystack.
Considerations for biomarker development. Drug
Discov. Today 12, 289294 (2007).
22. Little, J. D. C. A proof of the queuing formula L = W.
Oper. Res. 9, 383387 (1961).
23. Bunch, P. R. & Schacht, A. L. Modeling resource
requirements for pharmaceutical R.&D. Res. Technol.
Manage. 45, 4856 (2002).
24. Deloitte Touche Tohmatsu. Threading the talent needle:
what global executives are saying about people and
work. [online], http://www.deloitte.com/assets/
Dcom-Shared%20Assets/Documents/us_talent_
ThreadingtheTalentNeedle_final3409.pdf (2009).
25. PricewaterhouseCoopers. The changing dynamics of
pharma outsourcing in Asia: are you readjusting your
sights? [online], http://www.pwc.com/gx/en/
pharma-life-sciences/outsourcing/index.jhtml (2008).
26. Danzon, P. M., Nicholson S. & Pereira N. S.
Productivity in pharmaceutical-biotechnology R&D:
the role of experience and alliances. J. Health Econ.
24, 317339 (2005).
27. Gregson, N., Sparrowhawk, K., Mauskopf, J. & Paul, J.
Pricing medicines: theory and practice, challenges and
opportunities. Nature Rev. Drug Discov. 4, 121130
(2005).
28. Woodcock, J. The prospects for personalized
medicine in drug development and drug therapy.
29. DiMasi, J. A. The value of improving the productivity of
the drug development process: faster times and better
decisions. Pharmacoeconomics 20, 110 (2002).
30. Pyzdek, T. The Six Sigma Handbook: The complete
guide for greenbelts, blackbelts, and managers of all
levels. 2
nd
revised edition (McGraw-Hill, New York,
2003).
31. Goldratt, E. M. & Cox J. The goal: a process of ongoing
improvement. Third edition. (Gower Publishing,
Aldershot, 2004).
32. Maca, J., Bhattacharya, S., Dragalin, V, Gallo, P,
Krams, M. Adaptive seamless phase II/III designs
background, operational aspects, and examples.
Drug Inf. J. 4, 463473 (2006).
33. Kramer, J. A., Sagart, J. E. & Morris, D. L. The
application of discovery toxicology and pathology
towards the design of safer pharmaceutical lead
candidates. Nature Rev. Drug Discov. 6, 636649
(2007).
34. Hopkins, A. L. & Groom, C. R. The druggable genome.
35. Grenet, O. Significance of the human genome sequence
to drug discovery. Pharmacogenomics J. 1, 1112
(2001).
36. Horton, J. D., Cohen, J. C. & Hobbs, H. H. PCSK9:
a convertase that coordinates LDL catabolism.
J. Lipid Res. S172S177 (2009).
37. Abifadel, M. et al. Mutations in PCSK9 cause
autosomal dominant hypercholesterolemia.
Nature Genet. 34, 154156 (2003).
38. Cohen, J. C., Boerwinkle, E., Mosley, Jr., T. H. &
Hobbs, H. H. Sequence variations in PCSK9, low LDL,
and protection against coronary heart disease.
N. Engl. J. Med. 354, 12641272 (2006).
39. Cao, G., Qian, Y. W., Kowala, M. C. & Konrad, R. J.
Further LDL cholesterol lowering through targeting
PCSK9 for coronary artery disease. Endocrine,
Metabolic & Immune Disorders Drug Targets. 8,
238243 (2008).
40. Chan, J.C. et al. A proprotein convertase subtilisin/
kexin type 9 neutralizing antibody reduces serum
cholesterol in mice and nonhuman primates. Proc. Natl
Acad. Sci. USA 106, 98209825 (2009).
41. Shepherd, J. et.al. Effect of lowering LDL cholesterol
substantially below currently recommended levels in
patients with coronary heart disease and diabetes.
Diabetes Care 29, 12201226 (2006).
42. Drenth, J. P. H. & Waxman, S. G. Mutations in sodium-
channel gene SCN9A cause a spectrum of human
genetic pain disorders. J. Clin. Invest. 117,
36033609 (2007).
43. Krafte, D. S. & Bannon, A. W. Sodium channels and
nociception: recent concepts and therapeutic
opportunities. Curr. Opin. Pharmacol. 8, 5056 (2008).
44. Bonabeau, E. N., Bodick, N. & Armstrong, R. W.
A more rational approach to new-product
development. Harvard Bus. Rev. 86, 96102 (2008).
A review of Eli Lillys Chorus model of early drug
development and FIPNet strategy.
45. Cuatrecasas, P. Drug discovery in jeopardy. J. Clin.
Invest. 116, 28372842 (2006).
A frank analysis by one of the most successful R&D
leaders of problems that have fostered the current
innovation crisis.
Acknowledgements
The authors wish to acknowledge the seminal thought leader-
ship on the issue of R&D productivity that A. Bingham has
provided to the pharmaceutical industry. In addition, we wish
to thank J.S. Andersen and T. Mason for helping frame the
productivity concepts expressed in this manuscript, and G.
Pisano and J. DiMasi for helpful suggestions and comments.
Competing interests statement
The authors declare competing financial interests: see web
version for details.
DATABASES
UniProtKB: http://www.uniprot.org
PCsK9 | sCn9A
SUPPLEMENTARY INFORMATION
see online article: s1 (box) | s2 (box) | s3 (box)
All links Are AcTive in The online PDf
nrd_3078_mar10.indd 214 15/2/10 11:08:20
AnAlysi s
Cost per NME (US$ millions)
2,000 1,800 1,600 1,400 1,200 1,000 800 200 400 600 0
Cost per NME (baseline)
CT (lead optimizationPhase I)
C (Phase IIII)
p(TS) (shift from Phase II to Phase I)
Cost per NME (~28% reduction)
CT (Phase II, Phase III)
C (10% across all phases)
p(TS) (Phase III)
our Phase II portfolio to approximately 50%. Moreover,
the average C for establishing POc for a single NME in
chorus is $6 million compared with $22 million using
our more traditional approach. Thus, the epicentre of
R&D focus and investment is shifted from the tradi-
tional paradigm of resource-intensive Phase III and even
Phase Iv clinical trials to one that generates substantially
greater shots on goal in the form of high-quality drug
candidates (WIP) directed towards more validated tar-
gets and in which uncertainty surrounding these more
unprecedented targets (and mechanisms of drug action)
can be resolved earlier by POc studies.
Given that the vast majority of drug candidates are
destined to fail (even with reduced attrition) can they
fail faster and less expensively? This model obviously
requires a redistribution of R&D investments from the
later stages of development to the sweet spot illustrated in
FIG. 5. This can be achieved if late-phase attrition can be
reduced as we have described. For example, as previously
modelled, shifting attrition from Phase II or Phase III to
Phase I can reduce the cost of drug development
29
. Our
modelling shows that shifting 25% of Phase II attrition
to Phase I results in savings of $30 million in out-of-
pocket C per NME, which is enough to fund two addi-
tional Phase I WIPs. This equates to shifting Phase II
p(TS) from 34% to 41% and reducing Phase I p(TS)
from 54% to 45%. This cost saving would be multiplied
over an R&D portfolio that is aiming to produce more
than 1 NME per year and could be increased further if
some attrition was shifted from Phase III to Phase I (for
example, shifting 25% of Phase III attrition to Phase I
results in more than $20 million in additional out-of-
pocket savings).
however, one risk of using this quick win, fast fail
paradigm is that false-negative POc data (type 2 errors)
could offset some or all of the cost savings generated by
shifting attrition earlier
21
. Thus, it is important to ensure
that the additional studies that form the basis for the
early decision have a relatively low uncertainty, which is
not possible for all projects.
Finally, we consider that one of the major contribu-
tors to high Phase III attrition is simply premature
advancement of NMEs into Phase III; that is, prior to
establishing efficacy and an acceptable safety profile in
Phase I (POc) and Phase II. The latter is often the result
of poor discipline in portfolio management combined
with perceived near-term business imperatives.
Summary and additional considerations
without a substantial increase in R&D productivity, the
pharmaceutical industrys survival (let alone its contin-
ued growth prospects), at least in its current form, is in
great jeopardy. Much has been written on this subject
and multiple contributing factors and potential solutions
have been proposed
13,5
. Our econometric model of the
current cost of discovering and successfully develop-
ing a single new medicine reveals the major contribu-
tors to the escalating costs and the declining number of
launches. Importantly, our parametric analyses further
reveal where the greatest improvements in productivity
must occur.
we also offer a number of potential solutions, which
if effectively implemented could dramatically reduce
costs and importantly increase the flow of truly innova-
tive new medicines to patients. For example, using our
model parameters (FIG. 2) and sensitivity analyses (FIG. 3),
we can elaborate a series of interventions to reduce the
cost per NME by 28% and 50% respectively. FIG. 6 illus-
trates the potential interventions designed to increase
productivity by primarily reducing CT and C, as well as
by shifting attrition to early stages of clinical develop-
ment, increasing the overall p(TS) in Phase II and III.
Although the more extreme case of a 50% reduction in
cost per NME is ambitious, it is in our view not unre-
alistic and may be necessary to sustain a viable R&D
business model.
As the pharmaceutical industry transitions from
an era of me-too or slightly me-better drugs to one
of highly innovative medicines that result in mark-
edly improved health outcomes, it must, as our model
dictates, re-focus its resources (money and talent) on
discovery research and early translational medicine.
Although the scientific substrate for drug discovery has
never been more abundant, a more complete under-
standing of human (disease) biology will still be required
before many true breakthrough medicines emerge.
we also emphasize that there are other important
elements in a highly productive R&D enterprise that are
not included in our model, most of which are necessary,
although not sufficient to assure success. As underscored
by cuatrecasas
45
, fostering scientific creativity and being
opportunistic for serendipitous scientific and medical
findings are clearly important elements of past and future
success in the pharmaceutical industry. Over-managing or
Figure 6 | Productivity interventions yielding a substantial reduction in the cost
per new molecular entity. This analysis shows the potential beneficial effect on
capitalized cost per new molecular entity (nME) launch resulting from improvements
in a number of R&D parameters (cost, cycle time (CT), probability of technical success
(p(Ts))) as calculated using the R&D model described. A 28% reduction in cost per
nME could be achieved through the aggregate actions resulting in a 9% reduction in
CT from lead optimization to Phase I, a 10%, 15% and 20% reduction in cost for clinical
development Phases I, II and III respectively, and a shift in attrition from Phase II to
Phase I (decrease p(Ts) in Phase I by 17% and increase p(Ts) in Phase II by 47%).
An even larger 50% reduction in cost per nME could be achieved by additionally
reducing CT by 20% across Phases II and III, a further reduction in cost by 10% across
all phases, and an increase in Phase III p(Ts) by 14% without negatively affecting other
parameters.
nrd_3078_mar10.indd 213 15/2/10 11:08:20
AnAlysi s
even micro-managing R&D by those with little scientific
medical expertise or experience in R&D may well have
contributed to recent pipeline gaps and reduced R&D
productivity
45
. Moreover, fostering such scientific crea-
tivity and medical opportunism is often challenging, and
sometimes incompatible with the short-term and often
aspirational goals of business-driven enterprises. Most
importantly, in our opinion, there is no substitute for
scientific and clinical intuition for any successful R&D
organization, whether it be in the initial selection of drug
targets, the design of a clinical trial or the interpretation
of clinical data in order to advance a molecule into late-
phase clinical development. Simply stated, good process
will never be a substitute for good people or good science.
however, we firmly believe that both good process and
good science are not only compatible, but together will
yield the greatest return on R&D investments and thus,
have the greatest impact on R&D productivity.
1. OHagan, P. & Farkas, C. Bringing pharma R.&D back
to health. Bain Brief [online], http://www.bain.com/
bainweb/PDFs/cms/Public/BB_Managing_RandD_HC.
pdf (2009).
2. Lindgardt, Z., Reeves, M. & Wallenstein, J. Waking the
giant: business model innovation in the drug industry.
In Vivo 26, 16 (2008).
3. Garnier, J. Rebuilding the R.&D engine in big pharma.
Harvard Bus. Rev. 86, 6876 (2008).
A candid assessment by the former CEO of
GlaxoSmithKline of the downsides of the industrys
process culture.
4. Angell, M. The Truth about Drug Companies: How They
Deceive Us and What to do About It. (Random House
Trade Paperbacks, New York, 2005).
5. Chertkow, J. Trends in pharmaceutical portfolio
management: strategies to maintain profitability
despite adversity. Datamonitor [online] http://www.
datamonitor.com (2008).
6. Murray, A., Berndt, E. R. & Cutler, D. M. Prescription
drug spending trends in the United States: looking
beyond the turning point. Health Affairs w151w160,
(2008).
7. EvaluatePharma Alpha World Preview 2014. Evaluate
Pharma report (2009).
8. Goodman, M. Market watch: Pharma industry
performance metrics: 20072012E. Nature Rev. Drug
Discov. 7, 795 (2008).
9. Mathieu, M. P., ed. Parexels Bio/Pharmaceutical
R&D Statistical Sourcebook 2008/2009.
(Parexel International Corporation, Waltham, 2008).
10. Pisano, G. P. Science Business: the Promise, the
Reality, and the Future of Biotech. (Harvard Business
School Press, Boston, 2006).
An insightful retrospective on the biotech industry,
and the reasons why it has not yet lived up to its
promises.
11. DiMasi, J. A. & Grabowski, H. G. The cost of
biopharmaceutical R&D: Is biotech different?
Manage. Decis. Econ. 28, 469479 (2007).
A contemporary assessment of the costs to
discover, develop and launch new molecular entities.
12. Kola, I. The state of innovation in drug development.
13. Cohen, F. J. Macro trends in pharmaceutical
(2005).
14. Lichtenberg, F. R. The impact of new drug launches on
longevity: evidence from longitudinal, disease-level
data from 52 countries, 19822001 Int. J. Health
Care Finance Econ. 5, 4743 (2005).
15. Kimball, T. R., McCoy, C. E., Khoury, P. R., Daniels, S. R.
& Dolan, L. M. Obesity, diabetes damage young
arteries, could shorten life. News release, American
Heart Association website [online] http://
americanheart.mediaroom.com/index.
php?s=43&item=740 (2009).
16. KMR Group. Pharmaceutical Benchmarking Forum
[online] http://kmrgroup.com/ForumsPharma.html
(2009).
17. Booth, B. & Zemmil, R. Prospects for productivity.
Booth and Zemmil framed the issue of productivity
in the pharmaceutical industry and provide a
qualitative discussion of some of the drivers and
potential solutions.
18. Hu, M., Schultz, K., Sheu, J. & Tschopp, D. The
innovation gap in pharmaceutical drug discovery &
new models for R&D success. [online], http://www.
kellogg.northwestern.edu/academic/biotech/faculty/
articles/NewRDModel.pdf (2007).
711715 (2004).
This article provides an in-depth analysis of attrition
by phase and therapy area, and offers a scientific
approach to reduce Phase II and Phase III attrition
rates.
20. Innovation or Stagnation? Challenge and Opportunity
on the Critical Path to New Medical Products [online],
http://www.fda.gov/oc/initiatives/criticalpath/
whitepaper.html (2004).
A seminal paper on the hurdles that must be
overcome to reverse the decline in R&D productivity.
21. Peck, R. W. Driving earlier clinical attrition: if you want
to find the needle, burn down the haystack.
Considerations for biomarker development. Drug
Discov. Today 12, 289294 (2007).
22. Little, J. D. C. A proof of the queuing formula L = W.
Oper. Res. 9, 383387 (1961).
23. Bunch, P. R. & Schacht, A. L. Modeling resource
requirements for pharmaceutical R.&D. Res. Technol.
Manage. 45, 4856 (2002).
24. Deloitte Touche Tohmatsu. Threading the talent needle:
what global executives are saying about people and
work. [online], http://www.deloitte.com/assets/
Dcom-Shared%20Assets/Documents/us_talent_
ThreadingtheTalentNeedle_final3409.pdf (2009).
25. PricewaterhouseCoopers. The changing dynamics of
pharma outsourcing in Asia: are you readjusting your
sights? [online], http://www.pwc.com/gx/en/
pharma-life-sciences/outsourcing/index.jhtml (2008).
26. Danzon, P. M., Nicholson S. & Pereira N. S.
Productivity in pharmaceutical-biotechnology R&D:
the role of experience and alliances. J. Health Econ.
24, 317339 (2005).
27. Gregson, N., Sparrowhawk, K., Mauskopf, J. & Paul, J.
Pricing medicines: theory and practice, challenges and
opportunities. Nature Rev. Drug Discov. 4, 121130
(2005).
28. Woodcock, J. The prospects for personalized
medicine in drug development and drug therapy.
29. DiMasi, J. A. The value of improving the productivity of
the drug development process: faster times and better
decisions. Pharmacoeconomics 20, 110 (2002).
30. Pyzdek, T. The Six Sigma Handbook: The complete
guide for greenbelts, blackbelts, and managers of all
levels. 2
nd
revised edition (McGraw-Hill, New York,
2003).
31. Goldratt, E. M. & Cox J. The goal: a process of ongoing
improvement. Third edition. (Gower Publishing,
Aldershot, 2004).
32. Maca, J., Bhattacharya, S., Dragalin, V, Gallo, P,
Krams, M. Adaptive seamless phase II/III designs
background, operational aspects, and examples.
Drug Inf. J. 4, 463473 (2006).
33. Kramer, J. A., Sagart, J. E. & Morris, D. L. The
application of discovery toxicology and pathology
towards the design of safer pharmaceutical lead
candidates. Nature Rev. Drug Discov. 6, 636649
(2007).
34. Hopkins, A. L. & Groom, C. R. The druggable genome.
35. Grenet, O. Significance of the human genome sequence
to drug discovery. Pharmacogenomics J. 1, 1112
(2001).
36. Horton, J. D., Cohen, J. C. & Hobbs, H. H. PCSK9:
a convertase that coordinates LDL catabolism.
J. Lipid Res. S172S177 (2009).
37. Abifadel, M. et al. Mutations in PCSK9 cause
autosomal dominant hypercholesterolemia.
Nature Genet. 34, 154156 (2003).
38. Cohen, J. C., Boerwinkle, E., Mosley, Jr., T. H. &
Hobbs, H. H. Sequence variations in PCSK9, low LDL,
and protection against coronary heart disease.
N. Engl. J. Med. 354, 12641272 (2006).
39. Cao, G., Qian, Y. W., Kowala, M. C. & Konrad, R. J.
Further LDL cholesterol lowering through targeting
PCSK9 for coronary artery disease. Endocrine,
Metabolic & Immune Disorders Drug Targets. 8,
238243 (2008).
40. Chan, J.C. et al. A proprotein convertase subtilisin/
kexin type 9 neutralizing antibody reduces serum
cholesterol in mice and nonhuman primates. Proc. Natl
Acad. Sci. USA 106, 98209825 (2009).
41. Shepherd, J. et.al. Effect of lowering LDL cholesterol
substantially below currently recommended levels in
patients with coronary heart disease and diabetes.
Diabetes Care 29, 12201226 (2006).
42. Drenth, J. P. H. & Waxman, S. G. Mutations in sodium-
channel gene SCN9A cause a spectrum of human
genetic pain disorders. J. Clin. Invest. 117,
36033609 (2007).
43. Krafte, D. S. & Bannon, A. W. Sodium channels and
nociception: recent concepts and therapeutic
opportunities. Curr. Opin. Pharmacol. 8, 5056 (2008).
44. Bonabeau, E. N., Bodick, N. & Armstrong, R. W.
A more rational approach to new-product
development. Harvard Bus. Rev. 86, 96102 (2008).
A review of Eli Lillys Chorus model of early drug
development and FIPNet strategy.
45. Cuatrecasas, P. Drug discovery in jeopardy. J. Clin.
Invest. 116, 28372842 (2006).
A frank analysis by one of the most successful R&D
leaders of problems that have fostered the current
innovation crisis.
Acknowledgements
The authors wish to acknowledge the seminal thought leader-
ship on the issue of R&D productivity that A. Bingham has
provided to the pharmaceutical industry. In addition, we wish
to thank J.S. Andersen and T. Mason for helping frame the
productivity concepts expressed in this manuscript, and G.
Pisano and J. DiMasi for helpful suggestions and comments.
DATABASES
UniProtKB: http://www.uniprot.org
PCsK9 | sCn9A
see online article: s1 (box) | s2 (box) | s3 (box)
All links Are AcTive in The online PDf
nrd_3078_mar10.indd 214 15/2/10 11:08:20
AnAlysi s
Cost per NME (US$ millions)
2,000 1,800 1,600 1,400 1,200 1,000 800 200 400 600 0
Cost per NME (baseline)
CT (lead optimizationPhase I)
C (Phase IIII)
p(TS) (shift from Phase II to Phase I)
CT (Phase II, Phase III)
C (10% across all phases)
p(TS) (Phase III)
our Phase II portfolio to approximately 50%. Moreover,
the average C for establishing POc for a single NME in
chorus is $6 million compared with $22 million using
our more traditional approach. Thus, the epicentre of
R&D focus and investment is shifted from the tradi-
tional paradigm of resource-intensive Phase III and even
Phase Iv clinical trials to one that generates substantially
greater shots on goal in the form of high-quality drug
candidates (WIP) directed towards more validated tar-
gets and in which uncertainty surrounding these more
unprecedented targets (and mechanisms of drug action)
can be resolved earlier by POc studies.
Given that the vast majority of drug candidates are
destined to fail (even with reduced attrition) can they
fail faster and less expensively? This model obviously
requires a redistribution of R&D investments from the
later stages of development to the sweet spot illustrated in
FIG. 5. This can be achieved if late-phase attrition can be
reduced as we have described. For example, as previously
modelled, shifting attrition from Phase II or Phase III to
Phase I can reduce the cost of drug development
29
. Our
modelling shows that shifting 25% of Phase II attrition
to Phase I results in savings of $30 million in out-of-
pocket C per NME, which is enough to fund two addi-
tional Phase I WIPs. This equates to shifting Phase II
p(TS) from 34% to 41% and reducing Phase I p(TS)
from 54% to 45%. This cost saving would be multiplied
over an R&D portfolio that is aiming to produce more
than 1 NME per year and could be increased further if
some attrition was shifted from Phase III to Phase I (for
example, shifting 25% of Phase III attrition to Phase I
results in more than $20 million in additional out-of-
pocket savings).
however, one risk of using this quick win, fast fail
paradigm is that false-negative POc data (type 2 errors)
could offset some or all of the cost savings generated by
shifting attrition earlier
21
. Thus, it is important to ensure
that the additional studies that form the basis for the
early decision have a relatively low uncertainty, which is
not possible for all projects.
Finally, we consider that one of the major contribu-
tors to high Phase III attrition is simply premature
advancement of NMEs into Phase III; that is, prior to
establishing efficacy and an acceptable safety profile in
Phase I (POc) and Phase II. The latter is often the result
of poor discipline in portfolio management combined
with perceived near-term business imperatives.
Summary and additional considerations
without a substantial increase in R&D productivity, the
pharmaceutical industrys survival (let alone its contin-
ued growth prospects), at least in its current form, is in
great jeopardy. Much has been written on this subject
and multiple contributing factors and potential solutions
have been proposed
13,5
. Our econometric model of the
current cost of discovering and successfully develop-
ing a single new medicine reveals the major contribu-
tors to the escalating costs and the declining number of
launches. Importantly, our parametric analyses further
reveal where the greatest improvements in productivity
must occur.
we also offer a number of potential solutions, which
if effectively implemented could dramatically reduce
costs and importantly increase the flow of truly innova-
tive new medicines to patients. For example, using our
model parameters (FIG. 2) and sensitivity analyses (FIG. 3),
we can elaborate a series of interventions to reduce the
cost per NME by 28% and 50% respectively. FIG. 6 illus-
trates the potential interventions designed to increase
productivity by primarily reducing CT and C, as well as
by shifting attrition to early stages of clinical develop-
ment, increasing the overall p(TS) in Phase II and III.
Although the more extreme case of a 50% reduction in
cost per NME is ambitious, it is in our view not unre-
alistic and may be necessary to sustain a viable R&D
business model.
As the pharmaceutical industry transitions from
an era of me-too or slightly me-better drugs to one
of highly innovative medicines that result in mark-
edly improved health outcomes, it must, as our model
dictates, re-focus its resources (money and talent) on
discovery research and early translational medicine.
Although the scientific substrate for drug discovery has
never been more abundant, a more complete under-
standing of human (disease) biology will still be required
before many true breakthrough medicines emerge.
we also emphasize that there are other important
elements in a highly productive R&D enterprise that are
not included in our model, most of which are necessary,
although not sufficient to assure success. As underscored
by cuatrecasas
45
, fostering scientific creativity and being
opportunistic for serendipitous scientific and medical
findings are clearly important elements of past and future
success in the pharmaceutical industry. Over-managing or
Figure 6 | Productivity interventions yielding a substantial reduction in the cost
per new molecular entity. This analysis shows the potential beneficial effect on
capitalized cost per new molecular entity (nME) launch resulting from improvements
in a number of R&D parameters (cost, cycle time (CT), probability of technical success
(p(Ts))) as calculated using the R&D model described. A 28% reduction in cost per
nME could be achieved through the aggregate actions resulting in a 9% reduction in
CT from lead optimization to Phase I, a 10%, 15% and 20% reduction in cost for clinical
development Phases I, II and III respectively, and a shift in attrition from Phase II to
Phase I (decrease p(Ts) in Phase I by 17% and increase p(Ts) in Phase II by 47%).
An even larger 50% reduction in cost per nME could be achieved by additionally
reducing CT by 20% across Phases II and III, a further reduction in cost by 10% across
all phases, and an increase in Phase III p(Ts) by 14% without negatively affecting other
parameters.
nrd_3078_mar10.indd 213 15/2/10 11:08:20
AnAlysi s
multiple proteins may be required to perturb robust pheno-
types
13,16,17
. The emergent phenotype that occurs with the perturba-
tion of multiple nodes is experimentally demonstrated by synthetic
behaviors: synthetic lethality, synthetic sickness and synthetic rescue.
Dual knockouts in a number of model systems have shown that
although the isolated deletion of two individual genes may demon-
strate no effect, the simultaneous deletion of the two genes can be
lethal (synthetic lethality) or deleterious (synthetic sickness)
18
. A
recent large-scale study by Hillenmeyer et al. demonstrates the extent
of synthetic lethality when gene deletions are augmented by chemical
interventions
19
. Under ideal conditions, only 34% of single-gene
deletions in yeast resulted in lethality or sickness. However, when
the whole genome panel of yeast single-gene knockouts was screened
against a diverse small-molecule library and assayed against a
wide range of environmental conditions, an additional 63% of
gene knockouts showed a growth phenotype
19
, resulting in 97% of
genes demonstrating a tness defect when challenged with a
small molecule under at least one environmental condition. Thus,
although the majority of genes may be redundant under any
one environment, there seems to be little redundancy across a
spectrum of conditions when a genetic perturbation is combined
with a chemical insult.
As increased understanding of the role of networks in the robust-
ness and redundancy of biological systems challenges the dominant
assumption of single target drug discovery
16,2023
, a new approach to
drug discoverythat of polypharmacology
16,17,20,21,2429
is emer-
ging. Polypharmacology is not to be confused with the behavior of
promiscuous aggregators (as identied by McGovern et al.) that arise
from certain small molecules self-associating into colloids at
high concentrations in biological buffers
30
. Instead, polypharmacology
is the specic binding of a compound to two or more molecular
targets. In an analogy to Paracelsus axiom that the difference between
a drug and a poison is the dose, the utility or toxicity of synthetically
lethal and synthetically sick combinations is found in the biological
context to which they are applied. Therefore, understanding the
polypharmacology of a drug and its effect on biological networks
and phenotype is essential if we wish to improve efcacy but also
understand toxicity.
Synthetic lethality in cancer
The fundamental challenge of anticancer therapy is the need for agents
that eliminate cancer cells with a therapeutic index that is safely
tolerated by the patient. Most current anticancer drugs inhibit
essential functions that are present in both normal and cancerous
cells. Although these differentially impact rapidly dividing cancer cells,
the essential nature of the targets of most cytotoxic anticancer drugs
results in narrow therapeutic indices. In recent years, a new generation
of drugs have targeted protein kinases, such as ABL, EGRF and
ERBB2, that are differentially expressed in different cancers. These
new drugs, which target non-essential proteins, have more manageable
side effect proles than cytotoxics; however, clinical efcacy is, in
general, limited. An ideal cancer therapy, therefore, would be one that
targeted proteins or interactions that are essential in cancer cells but
non-essential in normal cells.
Cancer-specic molecular targets are rare. Most mutated oncogenic
proteins are also present in normal cells, and selective inhibition of the
mutant form can be a challenge. For example, the chronic myeloid
leukemia (CML)-specic BCR-ABL fusion protein is inhibited by
imatinib. However, imatinib also inhibits the non-oncogenic C-Abl
kinase in normal cells, and long-term administration of the drug can
lead to cardiotoxicity
31
. To overcome the difculty of identifying and
targeting differential features in a cancer, synthetic lethality has been
proposed as a possible strategy for therapeutic intervention
32
. In the
context of oncology, genetic and epigenetic changes in a cancer cell
may change not only the relative expression levels but also the
stoichiometry of the interaction network, and thus change the relative
dependence on specic proteins relative to normal cells. Thus, two
proteins that are non-essential in a normal cell may be essential in the
context of a re-wired cancer cell network. In short, though the
majority of the protein inventory in a cancer cell is the same as a
normal cell, the differences in the topology of the biological networks
could be targeted to produce an improved therapeutic index. Indeed,
subtle differences in network stability and structure between cancer
cells may explain the wide variance in cell fate that has been observed
in individual cells of the same genetic lineage
33
.
Whitehurst et al. recently conducted a whole-genome synthetic
lethality screen in combination with paclitaxel, resulting in the
discovery of new drug-drug combinations
34
. From the whole-genome
RNA interference screening, 87 initial genes were identied that
sensitized a human non-small-cell lung cancer line to paclitaxel,
including the gene encoding vacuolar ATPase, the target of salicyli-
halamide A. Subsequent testing of salicylihalamide A and paclitaxel in
combination was shown to reduce cancer cell viability. Sensitization
synthetic lethality screens can also be used to discover potential
synergistic combinations that can enhance the effectiveness of thera-
pies. For example, breast cancer cells with deciencies in BRAC1 and
BRAC2 show differential synthetic lethality to inhibition of poly(ADP-
ribose)-polymerase-1 (PARP). Screening a PARP inhibitor for addi-
tional synthetic lethality with an RNAi library identied a set of
kinases, including CDK5, whose knockdown resulted in increased
sensitization to the PARP inhibitor
35
. In addition to whole-genome
screening, hypotheses for new drug combinations can be discovered by
analysis of gene expression signatures. For example, analysis of breast
cancer gene expression data revealed that the gang of four (COX2,
MMP1, MMP2 and epiregulin) is essential for lung metastasis
36,37
.
Genetic and pharmacological inhibition of these four genes, in
combination, resulted in the halting of metastatic progression in a
mouse model
36
. Previous advances in cancer combination therapies,
such as those against childhood leukemia, were developed empirically
over three decades. Synthetic lethality chemical sensitization screens
offer a promising method to help systematically explore candidate
cancer drug combinations efciently in the laboratory.
The relationship between kinetics and systemic responses to per-
turbations offers an intriguing additional dimension in which network
pharmacology strategies can be applied; it also provides a framework
for understanding systems responses
3840
. The sequence in which a
combination is dosed may create different perturbations to the net-
work that may have a dramatic effect on efcacy
38,41
. The systemic
effects of network perturbations suggest that further studies on dosing
sequence should follow the discovery of even modest effects of
combination therapies.
Antibacterial polypharmacology
The single-target approach has been a major assumption behind
genomics-based drug discovery strategies so far, including the search
for new antibacterial targets
4245
. However, the biologically led strategy
for new antibacterial drugs, usually consisting of the search for single
proteins that are essential when deleted, is awed for two fundamental
reasons: the downstream difculty of discovering small-molecule
compounds has often been considered only after signicant invest-
ment in biology and a single amino acid mutation in the target protein
is often enough to confer drug resistance. Many effective antibiotics
RE VI E W
NATURE CHEMICAL BIOLOGY VOLUME 4 NUMBER 11 NOVEMBER 2008 683
Network pharmacology: the next paradigm in
drug discovery
Andrew L Hopkins
The dominant paradigm in drug discovery is the concept of designing maximally selective ligands to act on individual drug
targets. However, many effective drugs act via modulation of multiple proteins rather than single targets. Advances in systems
biology are revealing a phenotypic robustness and a network structure that strongly suggests that exquisitely selective compounds,
compared with multitarget drugs, may exhibit lower than desired clinical efcacy. This new appreciation of the role of
polypharmacology has signicant implications for tackling the two major sources of attrition in drug developmentefcacy and
toxicity. Integrating network biology and polypharmacology holds the promise of expanding the current opportunity space for
druggable targets. However, the rational design of polypharmacology faces considerable challenges in the need for new methods
to validate target combinations and optimize multiple structure-activity relationships while maintaining drug-like properties.
Advances in these areas are creating the foundation of the next paradigm in drug discovery: network pharmacology.
Over the past decade, there has been a signicant decrease in the rate
that new drug candidates are being translated into effective therapies in
the clinic. In particular, there has been a worrying rise in late-stage
attrition in phase 2 and phase 3 (ref. 1). Currently, the two single most
important reasons for attrition in clinical development are (i) lack of
efcacy and (ii) clinical safety or toxicology, which each account for
30% of failures
1
. These late-stage attrition rates are at the heart of much
of the relative decline in productivity of the pharmaceutical industry.
Moreover, the decline in productivity is creating a major nancial
shock to the pharmaceutical industry. Owing to patents expiring on
the current generation of marketed drugs, from 2010 onward, phar-
maceutical companies will face the rst fall in revenue in four decades.
Many reasons have been proposed for this decline in pharmaceu-
tical research and development productivity. However, the funda-
mental problem may not be technological, environmental or even
scientic but philosophicalthere may be issues with the core
assumptions that frame our approach to drug discovery. The increase
in the rate of drugs failing in late-stage clinical development over the
past decade has been concurrent with the dominance of the assump-
tion that the goal of drug discovery is to design exquisitely selective
ligands that act on a single disease target. This philosophy of rational
drug design, or more specically, the one gene, one drug, one disease
paradigm, arose from a congruence between genetic reductionism and
new molecular biology technologies that enabled the isolation and
characterization of individual disease-causing genes
2
, thereby
enabling the full realization of Ehrlichs philosophy of magic bullets
targeting individual chemoreceptors
3
. The underlying assumption of
the current approach is that safer, more effective drugs will result from
designing very selective ligands where undesirable and potentially
toxic side activities have been removed. However, after nearly two
decades of focusing on developing highly selective ligands, the clinical
attrition gures challenge this hypothesis.
Need for a one-two punch
Clinical attrition rates are not the only data to challenge the current
paradigm in drug discovery. Large-scale functional genomics studies
in a variety of model organisms have revealed that under laboratory
conditions, many single-gene knockouts by themselves exhibit little or
no effect on phenotype, with approximately 19% of genes being
essential across a number of model organisms
46
. In addition to the
19% lethality rate, systematic genome-wide homozygous gene deletion
experiments in yeast reveal that only 15% of knockouts result in a
tness defect in ideal conditions
7
. A project to delete each of the
druggable genes
8
in the mouse genome and prole each knockout
across a battery of phenotypic assays has revealed that as few as 10% of
knockouts demonstrate phenotypes that may be of value for drug
target validation
4,911
.
This robustness of phenotype can be understood in terms of
redundant functions and alternative compensatory signaling routes
12
.
Network analysis of biological pathways and interactions has revealed
that much of the robustness of biological systems can derive from the
structure of the network
13,14
. The scale-free nature of many biological
networks results in systems that are resilient against random deletion
of any one node but that are also critically dependent on a few highly
connected hubs. The inherent robustness of interaction networks, as
an underlying property, has profound implications for drug discovery;
instead of searching for the disease-causing genes, network biology
suggests that the strategy should be to identify the perturbations in the
disease-causing network
15
.
Network biology analysis predicts that if, in most cases, deletion of
individual nodes has little effect on disease networks, modulating Published online 20 October 2008; doi:10.1038/nchembio.118
Division of Biological Chemistry and Drug Discovery, College of Life Science,
University of Dundee, Dow Street, Dundee DD1 3DF, UK. Correspondence should
be addressed to A.L.H. (a.hopkins@dundee.ac.uk).
682 VOLUME 4 NUMBER 11 NOVEMBER 2008 NATURE CHEMICAL BIOLOGY
RE VI E W
First published in Nature Chemical Biology 4, 682 - 690 (2008)
multiple proteins may be required to perturb robust pheno-
types
13,16,17
. The emergent phenotype that occurs with the perturba-
tion of multiple nodes is experimentally demonstrated by synthetic
behaviors: synthetic lethality, synthetic sickness and synthetic rescue.
Dual knockouts in a number of model systems have shown that
although the isolated deletion of two individual genes may demon-
strate no effect, the simultaneous deletion of the two genes can be
lethal (synthetic lethality) or deleterious (synthetic sickness)
18
. A
recent large-scale study by Hillenmeyer et al. demonstrates the extent
of synthetic lethality when gene deletions are augmented by chemical
interventions
19
. Under ideal conditions, only 34% of single-gene
deletions in yeast resulted in lethality or sickness. However, when
the whole genome panel of yeast single-gene knockouts was screened
against a diverse small-molecule library and assayed against a
wide range of environmental conditions, an additional 63% of
gene knockouts showed a growth phenotype
19
, resulting in 97% of
genes demonstrating a tness defect when challenged with a
small molecule under at least one environmental condition. Thus,
although the majority of genes may be redundant under any
one environment, there seems to be little redundancy across a
spectrum of conditions when a genetic perturbation is combined
with a chemical insult.
As increased understanding of the role of networks in the robust-
ness and redundancy of biological systems challenges the dominant
assumption of single target drug discovery
16,2023
, a new approach to
drug discoverythat of polypharmacology
16,17,20,21,2429
is emer-
ging. Polypharmacology is not to be confused with the behavior of
promiscuous aggregators (as identied by McGovern et al.) that arise
from certain small molecules self-associating into colloids at
high concentrations in biological buffers
30
. Instead, polypharmacology
is the specic binding of a compound to two or more molecular
targets. In an analogy to Paracelsus axiom that the difference between
a drug and a poison is the dose, the utility or toxicity of synthetically
lethal and synthetically sick combinations is found in the biological
context to which they are applied. Therefore, understanding the
polypharmacology of a drug and its effect on biological networks
and phenotype is essential if we wish to improve efcacy but also
understand toxicity.
Synthetic lethality in cancer
The fundamental challenge of anticancer therapy is the need for agents
that eliminate cancer cells with a therapeutic index that is safely
tolerated by the patient. Most current anticancer drugs inhibit
essential functions that are present in both normal and cancerous
cells. Although these differentially impact rapidly dividing cancer cells,
the essential nature of the targets of most cytotoxic anticancer drugs
results in narrow therapeutic indices. In recent years, a new generation
of drugs have targeted protein kinases, such as ABL, EGRF and
ERBB2, that are differentially expressed in different cancers. These
new drugs, which target non-essential proteins, have more manageable
side effect proles than cytotoxics; however, clinical efcacy is, in
general, limited. An ideal cancer therapy, therefore, would be one that
targeted proteins or interactions that are essential in cancer cells but
non-essential in normal cells.
Cancer-specic molecular targets are rare. Most mutated oncogenic
proteins are also present in normal cells, and selective inhibition of the
mutant form can be a challenge. For example, the chronic myeloid
leukemia (CML)-specic BCR-ABL fusion protein is inhibited by
imatinib. However, imatinib also inhibits the non-oncogenic C-Abl
kinase in normal cells, and long-term administration of the drug can
lead to cardiotoxicity
31
. To overcome the difculty of identifying and
targeting differential features in a cancer, synthetic lethality has been
proposed as a possible strategy for therapeutic intervention
32
. In the
context of oncology, genetic and epigenetic changes in a cancer cell
may change not only the relative expression levels but also the
stoichiometry of the interaction network, and thus change the relative
dependence on specic proteins relative to normal cells. Thus, two
proteins that are non-essential in a normal cell may be essential in the
context of a re-wired cancer cell network. In short, though the
majority of the protein inventory in a cancer cell is the same as a
normal cell, the differences in the topology of the biological networks
could be targeted to produce an improved therapeutic index. Indeed,
subtle differences in network stability and structure between cancer
cells may explain the wide variance in cell fate that has been observed
in individual cells of the same genetic lineage
33
.
Whitehurst et al. recently conducted a whole-genome synthetic
lethality screen in combination with paclitaxel, resulting in the
discovery of new drug-drug combinations
34
. From the whole-genome
RNA interference screening, 87 initial genes were identied that
sensitized a human non-small-cell lung cancer line to paclitaxel,
including the gene encoding vacuolar ATPase, the target of salicyli-
halamide A. Subsequent testing of salicylihalamide A and paclitaxel in
combination was shown to reduce cancer cell viability. Sensitization
synthetic lethality screens can also be used to discover potential
synergistic combinations that can enhance the effectiveness of thera-
pies. For example, breast cancer cells with deciencies in BRAC1 and
BRAC2 show differential synthetic lethality to inhibition of poly(ADP-
ribose)-polymerase-1 (PARP). Screening a PARP inhibitor for addi-
tional synthetic lethality with an RNAi library identied a set of
kinases, including CDK5, whose knockdown resulted in increased
sensitization to the PARP inhibitor
35
. In addition to whole-genome
screening, hypotheses for new drug combinations can be discovered by
analysis of gene expression signatures. For example, analysis of breast
cancer gene expression data revealed that the gang of four (COX2,
MMP1, MMP2 and epiregulin) is essential for lung metastasis
36,37
.
Genetic and pharmacological inhibition of these four genes, in
combination, resulted in the halting of metastatic progression in a
mouse model
36
. Previous advances in cancer combination therapies,
such as those against childhood leukemia, were developed empirically
over three decades. Synthetic lethality chemical sensitization screens
offer a promising method to help systematically explore candidate
cancer drug combinations efciently in the laboratory.
The relationship between kinetics and systemic responses to per-
turbations offers an intriguing additional dimension in which network
pharmacology strategies can be applied; it also provides a framework
for understanding systems responses
3840
. The sequence in which a
combination is dosed may create different perturbations to the net-
work that may have a dramatic effect on efcacy
38,41
. The systemic
effects of network perturbations suggest that further studies on dosing
sequence should follow the discovery of even modest effects of
combination therapies.
Antibacterial polypharmacology
The single-target approach has been a major assumption behind
genomics-based drug discovery strategies so far, including the search
for new antibacterial targets
4245
. However, the biologically led strategy
for new antibacterial drugs, usually consisting of the search for single
proteins that are essential when deleted, is awed for two fundamental
reasons: the downstream difculty of discovering small-molecule
compounds has often been considered only after signicant invest-
ment in biology and a single amino acid mutation in the target protein
is often enough to confer drug resistance. Many effective antibiotics
RE VI E W
Network pharmacology: the next paradigm in
drug discovery
Andrew L Hopkins
The dominant paradigm in drug discovery is the concept of designing maximally selective ligands to act on individual drug
targets. However, many effective drugs act via modulation of multiple proteins rather than single targets. Advances in systems
biology are revealing a phenotypic robustness and a network structure that strongly suggests that exquisitely selective compounds,
compared with multitarget drugs, may exhibit lower than desired clinical efcacy. This new appreciation of the role of
polypharmacology has signicant implications for tackling the two major sources of attrition in drug developmentefcacy and
toxicity. Integrating network biology and polypharmacology holds the promise of expanding the current opportunity space for
druggable targets. However, the rational design of polypharmacology faces considerable challenges in the need for new methods
to validate target combinations and optimize multiple structure-activity relationships while maintaining drug-like properties.
Advances in these areas are creating the foundation of the next paradigm in drug discovery: network pharmacology.
Over the past decade, there has been a signicant decrease in the rate
that new drug candidates are being translated into effective therapies in
the clinic. In particular, there has been a worrying rise in late-stage
attrition in phase 2 and phase 3 (ref. 1). Currently, the two single most
important reasons for attrition in clinical development are (i) lack of
efcacy and (ii) clinical safety or toxicology, which each account for
30% of failures
1
. These late-stage attrition rates are at the heart of much
of the relative decline in productivity of the pharmaceutical industry.
Moreover, the decline in productivity is creating a major nancial
shock to the pharmaceutical industry. Owing to patents expiring on
the current generation of marketed drugs, from 2010 onward, phar-
maceutical companies will face the rst fall in revenue in four decades.
Many reasons have been proposed for this decline in pharmaceu-
tical research and development productivity. However, the funda-
mental problem may not be technological, environmental or even
scientic but philosophicalthere may be issues with the core
assumptions that frame our approach to drug discovery. The increase
in the rate of drugs failing in late-stage clinical development over the
past decade has been concurrent with the dominance of the assump-
tion that the goal of drug discovery is to design exquisitely selective
ligands that act on a single disease target. This philosophy of rational
drug design, or more specically, the one gene, one drug, one disease
paradigm, arose from a congruence between genetic reductionism and
new molecular biology technologies that enabled the isolation and
characterization of individual disease-causing genes
2
, thereby
enabling the full realization of Ehrlichs philosophy of magic bullets
targeting individual chemoreceptors
3
. The underlying assumption of
the current approach is that safer, more effective drugs will result from
designing very selective ligands where undesirable and potentially
toxic side activities have been removed. However, after nearly two
decades of focusing on developing highly selective ligands, the clinical
attrition gures challenge this hypothesis.
Need for a one-two punch
Clinical attrition rates are not the only data to challenge the current
paradigm in drug discovery. Large-scale functional genomics studies
in a variety of model organisms have revealed that under laboratory
conditions, many single-gene knockouts by themselves exhibit little or
no effect on phenotype, with approximately 19% of genes being
essential across a number of model organisms
46
. In addition to the
19% lethality rate, systematic genome-wide homozygous gene deletion
experiments in yeast reveal that only 15% of knockouts result in a
tness defect in ideal conditions
7
. A project to delete each of the
druggable genes
8
in the mouse genome and prole each knockout
across a battery of phenotypic assays has revealed that as few as 10% of
knockouts demonstrate phenotypes that may be of value for drug
target validation
4,911
.
This robustness of phenotype can be understood in terms of
redundant functions and alternative compensatory signaling routes
12
.
Network analysis of biological pathways and interactions has revealed
that much of the robustness of biological systems can derive from the
structure of the network
13,14
. The scale-free nature of many biological
networks results in systems that are resilient against random deletion
of any one node but that are also critically dependent on a few highly
connected hubs. The inherent robustness of interaction networks, as
an underlying property, has profound implications for drug discovery;
instead of searching for the disease-causing genes, network biology
suggests that the strategy should be to identify the perturbations in the
disease-causing network
15
.
Network biology analysis predicts that if, in most cases, deletion of
individual nodes has little effect on disease networks, modulating Published online 20 October 2008; doi:10.1038/nchembio.118
Division of Biological Chemistry and Drug Discovery, College of Life Science,
University of Dundee, Dow Street, Dundee DD1 3DF, UK. Correspondence should
be addressed to A.L.H. (a.hopkins@dundee.ac.uk).
RE VI E W
way drugs perturb networks compared to genetic deletions. Genetic
deletions completely remove all the interactions and functions of the
node from a network, whereas a drug may only partially ablate some
interactions
78
(such as decreasing the concentration of a metabolite)
but leave other links fully intact (such as protein-protein interactions).
In contrast, agonists may strengthen particular links in a network.
Thus, modeling attacks on links rather than nodes should provide a
closer model of drug action
71,78
. Furthermore, an essential point is
that only about 15% of any protein nodes in a network may be
chemically tractable with small-molecule drugs
8
; therefore, it is
necessary to map druggability and polypharmacology interactions
on integrated biological networks in order to identify the optimal
points of interaction for drug discovery.
Designer polypharmacology
Network analysis does not preclude the identication of individual
targets that by virtue of their position in a disease network could be
modulated to achieve a benecial clinical outcome. Indeed, the
theoretical work on synthetic rescue suggests that inactivation of
one node, for example, by mutation or environmental factors, could
be compensated for by the therapeutic inactivation of a second
node
15,79
. Indeed, in terms of synthetic lethality, oncogenic mutations
may themselves form half of a synthetically lethal pair, thereby
resulting in the need to pharmacologically inhibit only one target
34,35
.
However, if drug hunters are to embrace the wider opportunity
posed by network pharmacology, then what is required is a playbook
of polypharmacology strategies to design therapies that act on several
nodes in a disease network (Box 1). For good reason, medicinal
chemists have tried to decrease the off-target effects of drug candidates
to try to decrease the chances of off-target toxicities. Analysis of the
Bioprint database of the complete screening matrix of FDA-approved
drugs against approximately 200 assays reveals a strong relationship
between calculated lipophilicity (clogP) and low-afnity off-target
promiscuity
80
. The number of potential off-target activities appears
to double above the value clogP 3.75 (ref. 80). The goal of
polypharmacology is not to lazily introduce nonspecic promiscuity
into a compound by increasing the lipophilicity but to identify a
compound with a desired biological prole across multiple targets
whose combined modulation will perturb a disease state. Thus,
understanding the broader polypharmacology prole of a compound
and rationally modifying its proles should equally benet safety
pharmacology as well as disease efcacy.
Specic design strategies are required to balance a set of biological
activities in one compound. Morphy et al. have described the
continuum of design strategies medicinal chemists have traditionally
used to design drugs with multiple activities
81
. At one end of the
spectrum are conjugated ligands, which contain separate pharmaco-
poeia entities connected by a linker. Ligands designed by conjugating
two distinct pharmacophores are more likely to have high molecular
weight and less likely to have oral drug-like physicochemical proper-
ties
82
. At the other end of the spectrum are ligands where multiple
pharmacophores overlap or are highly integrated. Compounds with
overlapping or integrated pharmacophores are likely to have lower
molecular weight and potentially more drug-like physicochemical
properties. Ligands where polypharmacology has been deliberately
designed in, by conjugation or overlapping pharmacophores, tend to
have lower ligand efciency than general preclinical compounds. This
nding is not surprising given that the compounds are not optimized
for one single target.
Designers of polypharmacology can also take lessons from the work
on drug resistance, particularly in the eld of HIV-1 therapies.
Mutations of single amino acid residues are often sufcient to confer
drug resistance to many anti-HIV drugs, such as the non-nucleoside
reverse transcriptase inhibitors (NNRTIs). Efforts to discover second-
generation NNRTIs required compounds to be active against both the
wild type and the common drug-resistant mutations. Thus, second-
generation NNRTIs can be considered as multitarget drugs as they are
required to bind to several structurally distinct binding sites on a
spectrum of mutated reverse transcriptases. Crystallographic analyses
of the NNRTIs revealed that one design strategy is to identify
inhibitors that make strong molecular interactions with conserved
regions of the binding site, such as structurally important residues and
main chain atoms; this has the added advantage of reducing depen-
dence on interactions with mutable residues
83
. Several of these design
strategies have also been derived a priori by theoretical analysis of
drug-resistant target ensembles
53
.
In addition to the relationship between lipophilicity and promis-
cuity, a strong relationship has also been reported between molecular
weight
25,82
and molecular complexity
84
and promiscuity. In an
Three strategies are available to the designers of multitarget therapies. The rst
strategy, which is the most conventional, is to prescribe multiple individual
medications. Multidrug combination cocktails are the mainstay of highly active
antiretroviral therapy for HIV and a large number of anticancer protocols. The
drawback of prescribing multiple medications is patient compliance and the
danger of drug-drug interactions. To overcome these issues, a second strategy is
the development of multicomponent drugs that contain two or more active
ingredients formulated in the same delivery device, such as a single pill, capsule
or inhaler
22,52
. Several successful drug combinations have now been reformu-
lated in single multicomponent medicines, such as Atripla, Advair, Caduet,
Combivir, Epzicom, Rebetron and Truvada. Advances in formulation technolo-
gies are expanding the number of drug combinations that can be effectively
combined into a single delivery mechanism. However, given the signicant
differences in pharmacokinetics, metabolisms and bioavailability, reformulation
of drug combinations is not a trivial problem. Further, two drugs that are
generally safe when dosed individually cannot be assumed to be safe in
combination. In addition to the possibility of adverse drug-drug interactions, if
the theory of network pharmacology indicates that an effect on phenotype may
derive from hitting multiple targets, then that combined phenotypic perturbation
may be efcacious or deleterious. The major challenge to both drug combination
strategies is the regulatory requirement for each individual drug to be shown to
be safe as an individual agent and in combination. Therefore, most multi-
component drug development has focused on exploiting combinations of
approved drugs, with the notable exception of the torcetrapib-atorvastatin
combination that was being developed by Pzer. A drawback of this approach
to drug combinations is that the target universe of the current pharmacopeia is
very limited. Current estimates are that the entire formulary of approximately
1,200 FDA-approved drugs only acts on about 320 molecular targets
115
(excluding the incredibly promiscuous kinase drug sunitinib (Sutent,
SU11248), which itself binds to 79 protein kinases with K
d
o 10 mM)
116
.
The third strategy for multitarget therapy is to design a single compound with
selective polypharmacology
25,27,81
. The advent of wide-ligand proling has
revealed the extent of polypharmacology across the pharmacopeia, and it has
also shown that many approved drugs act on multiple targets
25,115
. Dosing with a
single compound may have advantages over a drug combination in terms of
equitable pharmacokinetics and biodistribution. Indeed, troughs in drug expo-
sure due to incompatible pharmacokinetics between components of a combina-
tion therapy may create a low-dose window of opportunity where a reduced
selection pressure can lead to drug resistance. In terms of drug registration,
approval of a single compound acting on multiple targets faces signicantly lower
regulatory barriers than approval of a combination of new drugs.
Box 1 Polypharmacology playbook
RE VI E W
act by targeting multiple proteins simultaneously rather than indivi-
dual proteins
46
. For example, the antibacterial action of
b-lactams is dependent on the inhibition of at least two of the
multiple penicillin-binding proteins (PBPs). Indeed, because multiple
PBPs can be deleted with no effect on phenotype
47
, the strategy of
single target essentiality would not have discovered this important
class of antibacterial drug targets. Similarly, uoroquinolone antibio-
tics are dual-targeted inhibitors of the proteins ParC and GyrA
(ref. 48). D-Cycloserine inhibits four targets, both pairs of alanine
racemases and D-Ala-D-Ala ligases. Likewise, fosfomycin overcomes
the redundancy of UDP N-acetylglucosamine enolpyruvyl transferases
by inhibiting them both. Therefore, if we wish to design single drugs
that limit drug resistance, we could consider the development of
methods to search and prioritize which combinations of targets can be
inhibited by the same drug and are essential, either individually (dual
essentials) or in combination (synthetic lethals).
A strategy of antibacterial polypharmacology can challenge the
current approach to genome-based drug discovery of anti-infectives
in four important ways. First, the druggability of a target is
prioritized over single target essentiality. Second, the target does
not need to be unique to the organism or absent in the host.
Although many essential housekeeping enzymes may be common
between the host and infectious agent, drug selectivity between the
host and infectious agent can be determined at the binding site
level. Third, targets are sought that are lethal in combination but
may have been overlooked as non-essential in individual gene
knockout studies, and fourth, groups of targets that are predicted to
potentially bind the same compound are prioritized over singleton
druggable targets. By targeting two or more essential genes with a
single chemical agent, the ability to delay drug resistance is designed
into the target discovery strategy from the start. Given the failure of
current genome-based strategies for discovering new antibacterial
drugs
45
, learning the lessons of the previous successful generation of
antibacterial drugs may encourage the development of antibacterial
polypharmacology discovery strategies.
Topology of targets
The two key challenges facing the development of network pharma-
cology are identifying a node or combination of nodes in a biological
network whose perturbation results in a desired therapeutic out-
come
49,50
, and discovering agents with the desired polypharmacology
prole to perturb those nodes. Three complementary methods for the
comparative analysis of disease networks are systematic screening,
knowledge-based combinations and network analysis. Presently, com-
bination screening of mixtures of drugs, chemical tools and RNAi in
cell-based disease models is the most efcient means of systematically
discovering new drug-drug combinations and synthetically lethal gene
pairs. However, as with all preclinical models, active combinations
discovered in the laboratory do not necessarily translate into the
clinic
51
. Moreover, idealized synergist screens between drugs or RNAi
in preclinical assays do not account for the complications of dosing,
scheduling pharmacokinetics and metabolism necessary to optimize a
therapeutic drug cocktail in the clinic. Owing to the size of the search
matrix, RNAichemical sensitization screens are usually performed
with a single drug against a whole-genome RNAi array
34
. Likewise,
systematic combination screening of approximately 1,000 US Food
and Drug Administration (FDA)-approved drugs required the use of
high-throughput screening methods to assay the massive data matrix
required for the factorial isobolgram analysis of each combination
through the full dose range
52,53
. Therefore, if we are to effectively
search for new drug combinations, there is a pressing need for
computational methods that could reduce the global search space
for target combinations
49,51
.
Owing to the vast number of possible drug and target combinations
and the ethical considerations and resource constraints on using
in vivo models and conducting clinical trials, most new combinations
have been selected for empirical testing based on a knowledge of the
underlying disease biology. Such knowledge-based approaches are
often incremental but can make a dramatic impact on disease out-
comes, as demonstrated by the success of multidrug highly active
antiretroviral therapy in decreasing human immunodeciency virus
(HIV) mortality rates in the developed world. Advances in pathway
analysis
54
and text mining of the biomedical literature
5558
can
potentially be used to enable the large-scale text mining of disease
knowledge to postulate new combination hypotheses by associative
techniques of inductive and abductive inference
59,60
. However, though
advances in informatics methods can aid the generation of new
hypotheses from connecting concepts in the literature, the drawback
of the knowledge-based methods is that they do not provide a robust
modeling analysis of the emergent properties of a network and system
when perturbed in new ways. Thus, counterintuitive, paradoxical and
unexpected system responses cannot necessarily be predicted by these
associative methods
39
.
An intriguing possibility for systematic target identication is that
the structures of biological networks themselves may provide valuable
information in assessing targets and their combinations
14,17,38,61
. Early
network analysis indicated the possibility of a direct correlation
between lethality and the degree of connectivity of nodes, where highly
connected hubs in protein interaction networks are more likely to be
essential
62
. Subsequent re-analysis of the data challenged the relation-
ship between the number of interactions of a protein and its essenti-
ality
63
. However, the hypothesis that protein function relates to
network topology has been strengthened by recent work that has
rened the relationship between network topology and system function
by focusing on betweenness centrality (the number of nonredundant
shortest paths traveling through a node
64,65
) and bridging centrality
(nodes between and connecting subgraph clusters dened by the ratio
of the number of interactions of a neighboring node in a subgraph over
the number of remaining edges in the subgraph
66
), in addition to the
metric of the degree centrality (the number of direct interactions
intersecting a node
14
). Bottlenecks with high betweenness values tend
to be better correlated with gene expression dynamics and essentiality
than highly connected hubs
64,67
. These ndings complement the
analysis on party and dates hubs, which suggests that hubs with
high betweenness values have pleiotropic functions across the net-
work
68
. Unexpectedly, non-hub bottlenecks with transient interac-
tions
65
and bridging proteins are less likely to be lethal than average
and tend to be independently regulated
66
. Thus, given their position in
communication between network clusters and their low lethality,
bridging nodes have been suggested as potential drug targets, although
modulation of the bridging targets themselves may still be indirect
66
.
An initial network analysis on the current drug targets of approved
drugs indicated that drug targets are commonly highly connected but
not essential
69,70
.
Despite initial challenges, several compelling theoretical and experi-
mental studies support the hypothesis that network topology is an
essential feature in the emergent system function of the protein when
it is perturbed; thus, this gives hope that systematic network analyses
may be a useful basis for developing methods to prioritize drug targets
and combinations of targets
17,61,7177
. However, if we are to success-
fully exploit network analysis for target identication, then we must
recognize the fundamentally different dynamics and kinetics in the
RE VI E W
way drugs perturb networks compared to genetic deletions. Genetic
deletions completely remove all the interactions and functions of the
node from a network, whereas a drug may only partially ablate some
interactions
78
(such as decreasing the concentration of a metabolite)
but leave other links fully intact (such as protein-protein interactions).
In contrast, agonists may strengthen particular links in a network.
Thus, modeling attacks on links rather than nodes should provide a
closer model of drug action
71,78
. Furthermore, an essential point is
that only about 15% of any protein nodes in a network may be
chemically tractable with small-molecule drugs
8
; therefore, it is
necessary to map druggability and polypharmacology interactions
on integrated biological networks in order to identify the optimal
points of interaction for drug discovery.
Designer polypharmacology
Network analysis does not preclude the identication of individual
targets that by virtue of their position in a disease network could be
modulated to achieve a benecial clinical outcome. Indeed, the
theoretical work on synthetic rescue suggests that inactivation of
one node, for example, by mutation or environmental factors, could
be compensated for by the therapeutic inactivation of a second
node
15,79
. Indeed, in terms of synthetic lethality, oncogenic mutations
may themselves form half of a synthetically lethal pair, thereby
resulting in the need to pharmacologically inhibit only one target
34,35
.
However, if drug hunters are to embrace the wider opportunity
posed by network pharmacology, then what is required is a playbook
of polypharmacology strategies to design therapies that act on several
nodes in a disease network (Box 1). For good reason, medicinal
chemists have tried to decrease the off-target effects of drug candidates
to try to decrease the chances of off-target toxicities. Analysis of the
Bioprint database of the complete screening matrix of FDA-approved
drugs against approximately 200 assays reveals a strong relationship
between calculated lipophilicity (clogP) and low-afnity off-target
promiscuity
80
. The number of potential off-target activities appears
to double above the value clogP 3.75 (ref. 80). The goal of
polypharmacology is not to lazily introduce nonspecic promiscuity
into a compound by increasing the lipophilicity but to identify a
compound with a desired biological prole across multiple targets
whose combined modulation will perturb a disease state. Thus,
understanding the broader polypharmacology prole of a compound
and rationally modifying its proles should equally benet safety
pharmacology as well as disease efcacy.
Specic design strategies are required to balance a set of biological
activities in one compound. Morphy et al. have described the
continuum of design strategies medicinal chemists have traditionally
used to design drugs with multiple activities
81
. At one end of the
spectrum are conjugated ligands, which contain separate pharmaco-
poeia entities connected by a linker. Ligands designed by conjugating
two distinct pharmacophores are more likely to have high molecular
weight and less likely to have oral drug-like physicochemical proper-
ties
82
. At the other end of the spectrum are ligands where multiple
pharmacophores overlap or are highly integrated. Compounds with
overlapping or integrated pharmacophores are likely to have lower
molecular weight and potentially more drug-like physicochemical
properties. Ligands where polypharmacology has been deliberately
designed in, by conjugation or overlapping pharmacophores, tend to
have lower ligand efciency than general preclinical compounds. This
nding is not surprising given that the compounds are not optimized
for one single target.
Designers of polypharmacology can also take lessons from the work
on drug resistance, particularly in the eld of HIV-1 therapies.
Mutations of single amino acid residues are often sufcient to confer
drug resistance to many anti-HIV drugs, such as the non-nucleoside
reverse transcriptase inhibitors (NNRTIs). Efforts to discover second-
generation NNRTIs required compounds to be active against both the
wild type and the common drug-resistant mutations. Thus, second-
generation NNRTIs can be considered as multitarget drugs as they are
required to bind to several structurally distinct binding sites on a
spectrum of mutated reverse transcriptases. Crystallographic analyses
of the NNRTIs revealed that one design strategy is to identify
inhibitors that make strong molecular interactions with conserved
regions of the binding site, such as structurally important residues and
main chain atoms; this has the added advantage of reducing depen-
dence on interactions with mutable residues
83
. Several of these design
strategies have also been derived a priori by theoretical analysis of
drug-resistant target ensembles
53
.
In addition to the relationship between lipophilicity and promis-
cuity, a strong relationship has also been reported between molecular
weight
25,82
and molecular complexity
84
and promiscuity. In an
Three strategies are available to the designers of multitarget therapies. The rst
strategy, which is the most conventional, is to prescribe multiple individual
medications. Multidrug combination cocktails are the mainstay of highly active
antiretroviral therapy for HIV and a large number of anticancer protocols. The
drawback of prescribing multiple medications is patient compliance and the
danger of drug-drug interactions. To overcome these issues, a second strategy is
the development of multicomponent drugs that contain two or more active
ingredients formulated in the same delivery device, such as a single pill, capsule
or inhaler
22,52
. Several successful drug combinations have now been reformu-
lated in single multicomponent medicines, such as Atripla, Advair, Caduet,
Combivir, Epzicom, Rebetron and Truvada. Advances in formulation technolo-
gies are expanding the number of drug combinations that can be effectively
combined into a single delivery mechanism. However, given the signicant
differences in pharmacokinetics, metabolisms and bioavailability, reformulation
of drug combinations is not a trivial problem. Further, two drugs that are
generally safe when dosed individually cannot be assumed to be safe in
combination. In addition to the possibility of adverse drug-drug interactions, if
the theory of network pharmacology indicates that an effect on phenotype may
derive from hitting multiple targets, then that combined phenotypic perturbation
may be efcacious or deleterious. The major challenge to both drug combination
strategies is the regulatory requirement for each individual drug to be shown to
be safe as an individual agent and in combination. Therefore, most multi-
component drug development has focused on exploiting combinations of
approved drugs, with the notable exception of the torcetrapib-atorvastatin
combination that was being developed by Pzer. A drawback of this approach
to drug combinations is that the target universe of the current pharmacopeia is
very limited. Current estimates are that the entire formulary of approximately
1,200 FDA-approved drugs only acts on about 320 molecular targets
115
(excluding the incredibly promiscuous kinase drug sunitinib (Sutent,
SU11248), which itself binds to 79 protein kinases with K
d
o 10 mM)
116
.
The third strategy for multitarget therapy is to design a single compound with
selective polypharmacology
25,27,81
. The advent of wide-ligand proling has
revealed the extent of polypharmacology across the pharmacopeia, and it has
also shown that many approved drugs act on multiple targets
25,115
. Dosing with a
single compound may have advantages over a drug combination in terms of
equitable pharmacokinetics and biodistribution. Indeed, troughs in drug expo-
sure due to incompatible pharmacokinetics between components of a combina-
tion therapy may create a low-dose window of opportunity where a reduced
selection pressure can lead to drug resistance. In terms of drug registration,
approval of a single compound acting on multiple targets faces signicantly lower
regulatory barriers than approval of a combination of new drugs.
Box 1 Polypharmacology playbook
RE VI E W
act by targeting multiple proteins simultaneously rather than indivi-
dual proteins
46
. For example, the antibacterial action of
b-lactams is dependent on the inhibition of at least two of the
multiple penicillin-binding proteins (PBPs). Indeed, because multiple
PBPs can be deleted with no effect on phenotype
47
, the strategy of
single target essentiality would not have discovered this important
class of antibacterial drug targets. Similarly, uoroquinolone antibio-
tics are dual-targeted inhibitors of the proteins ParC and GyrA
(ref. 48). D-Cycloserine inhibits four targets, both pairs of alanine
racemases and D-Ala-D-Ala ligases. Likewise, fosfomycin overcomes
the redundancy of UDP N-acetylglucosamine enolpyruvyl transferases
by inhibiting them both. Therefore, if we wish to design single drugs
that limit drug resistance, we could consider the development of
methods to search and prioritize which combinations of targets can be
inhibited by the same drug and are essential, either individually (dual
essentials) or in combination (synthetic lethals).
A strategy of antibacterial polypharmacology can challenge the
current approach to genome-based drug discovery of anti-infectives
in four important ways. First, the druggability of a target is
prioritized over single target essentiality. Second, the target does
not need to be unique to the organism or absent in the host.
Although many essential housekeeping enzymes may be common
between the host and infectious agent, drug selectivity between the
host and infectious agent can be determined at the binding site
level. Third, targets are sought that are lethal in combination but
may have been overlooked as non-essential in individual gene
knockout studies, and fourth, groups of targets that are predicted to
potentially bind the same compound are prioritized over singleton
druggable targets. By targeting two or more essential genes with a
single chemical agent, the ability to delay drug resistance is designed
into the target discovery strategy from the start. Given the failure of
current genome-based strategies for discovering new antibacterial
drugs
45
, learning the lessons of the previous successful generation of
antibacterial drugs may encourage the development of antibacterial
polypharmacology discovery strategies.
Topology of targets
The two key challenges facing the development of network pharma-
cology are identifying a node or combination of nodes in a biological
network whose perturbation results in a desired therapeutic out-
come
49,50
, and discovering agents with the desired polypharmacology
prole to perturb those nodes. Three complementary methods for the
comparative analysis of disease networks are systematic screening,
knowledge-based combinations and network analysis. Presently, com-
bination screening of mixtures of drugs, chemical tools and RNAi in
cell-based disease models is the most efcient means of systematically
discovering new drug-drug combinations and synthetically lethal gene
pairs. However, as with all preclinical models, active combinations
discovered in the laboratory do not necessarily translate into the
clinic
51
. Moreover, idealized synergist screens between drugs or RNAi
in preclinical assays do not account for the complications of dosing,
scheduling pharmacokinetics and metabolism necessary to optimize a
therapeutic drug cocktail in the clinic. Owing to the size of the search
matrix, RNAichemical sensitization screens are usually performed
with a single drug against a whole-genome RNAi array
34
. Likewise,
systematic combination screening of approximately 1,000 US Food
and Drug Administration (FDA)-approved drugs required the use of
high-throughput screening methods to assay the massive data matrix
required for the factorial isobolgram analysis of each combination
through the full dose range
52,53
. Therefore, if we are to effectively
search for new drug combinations, there is a pressing need for
computational methods that could reduce the global search space
for target combinations
49,51
.
Owing to the vast number of possible drug and target combinations
and the ethical considerations and resource constraints on using
in vivo models and conducting clinical trials, most new combinations
have been selected for empirical testing based on a knowledge of the
underlying disease biology. Such knowledge-based approaches are
often incremental but can make a dramatic impact on disease out-
comes, as demonstrated by the success of multidrug highly active
antiretroviral therapy in decreasing human immunodeciency virus
(HIV) mortality rates in the developed world. Advances in pathway
analysis
54
and text mining of the biomedical literature
5558
can
potentially be used to enable the large-scale text mining of disease
knowledge to postulate new combination hypotheses by associative
techniques of inductive and abductive inference
59,60
. However, though
advances in informatics methods can aid the generation of new
hypotheses from connecting concepts in the literature, the drawback
of the knowledge-based methods is that they do not provide a robust
modeling analysis of the emergent properties of a network and system
when perturbed in new ways. Thus, counterintuitive, paradoxical and
unexpected system responses cannot necessarily be predicted by these
associative methods
39
.
An intriguing possibility for systematic target identication is that
the structures of biological networks themselves may provide valuable
information in assessing targets and their combinations
14,17,38,61
. Early
network analysis indicated the possibility of a direct correlation
between lethality and the degree of connectivity of nodes, where highly
connected hubs in protein interaction networks are more likely to be
essential
62
. Subsequent re-analysis of the data challenged the relation-
ship between the number of interactions of a protein and its essenti-
ality
63
. However, the hypothesis that protein function relates to
network topology has been strengthened by recent work that has
rened the relationship between network topology and system function
by focusing on betweenness centrality (the number of nonredundant
shortest paths traveling through a node
64,65
) and bridging centrality
(nodes between and connecting subgraph clusters dened by the ratio
of the number of interactions of a neighboring node in a subgraph over
the number of remaining edges in the subgraph
66
), in addition to the
metric of the degree centrality (the number of direct interactions
intersecting a node
14
). Bottlenecks with high betweenness values tend
to be better correlated with gene expression dynamics and essentiality
than highly connected hubs
64,67
. These ndings complement the
analysis on party and dates hubs, which suggests that hubs with
high betweenness values have pleiotropic functions across the net-
work
68
. Unexpectedly, non-hub bottlenecks with transient interac-
tions
65
and bridging proteins are less likely to be lethal than average
and tend to be independently regulated
66
. Thus, given their position in
communication between network clusters and their low lethality,
bridging nodes have been suggested as potential drug targets, although
modulation of the bridging targets themselves may still be indirect
66
.
An initial network analysis on the current drug targets of approved
drugs indicated that drug targets are commonly highly connected but
not essential
69,70
.
Despite initial challenges, several compelling theoretical and experi-
mental studies support the hypothesis that network topology is an
essential feature in the emergent system function of the protein when
it is perturbed; thus, this gives hope that systematic network analyses
may be a useful basis for developing methods to prioritize drug targets
and combinations of targets
17,61,7177
. However, if we are to success-
fully exploit network analysis for target identication, then we must
recognize the fundamentally different dynamics and kinetics in the
RE VI E W
opportunity universe for single-target drugs (which is outlined by the
druggable genome
8
), not all target pair combinations will be accessible
to a single agent with drug-like properties. Predicting chemically
tractable combinations can increase the chances of nding a multi-
target compound. In order to develop a drug with a desired poly-
pharmacology prole, two problems need to be solved. First, a lead
compound needs to be identied with the desired biological activity
against multiple targets; then, this lead needs to be optimized into a
clinical candidate that combines the desired polypharmacology with a
safe, drug-like pharmaceutical prole.
The integration of in silico methods, combined with wide-ligand
biological proling against protein assays and gene expression arrays,
can provide drug designers with a new toolbox with which to assess
polypharmacology. First, proteins can be related by binding exact or
similar endogenous ligands or proteins; this can be determined by
exploiting data in gene ontologies and metabolic databases such as
KEGG
94,95
. Second, proteins can be related by known observed
polypharmacology of large sets of ligands such as those found in
pharmaceutical company screening sets and the medicinal chemistry
literature. A chemical network can be created that relates proteins that
bind the same ligand, where the strength of each edge can be
calculated using metrics of promiscuity
69,70,90,9699
.
If experimentally derived ligand data are not available, proteins can
be related by predictions of polypharmacology
90,100104
. Sets of ligands
for each protein obtained from chemogenomic data sources can be
used to train machine learning algorithms to predict pharmacology
activity proles. Bayesian approaches can be used to classify chemical
structures based on chemical ngerprints
90,100,101
. Dynamically
cross-comparing the chemical similarity of sets of ligands for each
protein is a complementary and promising method of predicting
polypharmacology proles from chemical structure
100,101
. The struc-
ture of the observed polypharmacology network itself also
provides information beyond chemical structure similarity that
could also be utilized using Bayesian network methods, as has
been the case in protein-protein interaction networks
105
. Though
promising, chemical ngerprintbased similarity methods paradoxi-
cally are incompatible with the two simplest relationshipsthose
of lipophilicity and molecular weight with promiscuityas larger
compounds tend to exhibit a greater number of ngerprints. The
development of improved polypharmacology prediction methods
will be an important topic in cheminformatics and toxicology.
Complementary to chemogenomics approaches to predict drug-
target relationships is the use of phenotypic data (such as clinical
side effects)
104,106
or gene expression proles
107
to cluster chemical
structures by functional effects. Clustering compounds by phenotypic
and gene signature proles enables unknown mechanisms to
be inferred.
Structural bioinformatics methods that map gene family sequence
alignments onto the structural information of protein binding sites
provide a valuable rst-order assessment of likely selectivity and
promiscuity within a gene family
95,108110
. Comparing protein binding
site similarity with the observed polypharmacology networks could
provide insight into protein binding site parameters that map with the
behavior of chemical networks. For example, analysis of the protein
kinase superfamily in the human genome reveals discrete clusters of
subfamilies when the full-length sequences are compared (Fig. 3).
However, these subfamily clusters break down when the sequence
similarity is measured at the level of the ATP binding site, where most
kinase inhibitors bind. Hence, at a rst approximation the difculties
drug designers have in ne tuning the selectivity proles of compe-
titive kinase inhibitors is not surprising
14,111,112
. Despite the inherent
challenge of achieving nely tuned kinase activity proles, deriving
kinase SAR data
112
from the wealth of large-scale chemogenomics
analyses and using subtle modication of existing kinase scaffolds to
exploit de-hydrons
111
are two promising strategies. Finally, if
three-dimensional protein structures of the targets of interest are
available, then parallel large-scale multitarget virtual screening is
also a promising method
113
.
Figure 2 Expanding opportunity for drug
discovery space with polypharmacology. A subset
of the network data shown in Figure 1 for
literature targets associated with asthma. Drug
targets are represented as nodes, and chemical
matter that binds to two or more nodes is
represented as edges. Targets are colored by gene
family. The color of the edges represents the
strength of the chemical network between two
targets as dened by the number of shared
compounds that are active against both targets
below an afnity of 1 mM: light blue (1 to
10 compounds) to black (41,000 compounds).
Of the 44 targets described in the literature
as potential drug targets for the treatment of
asthma, 44 share polypharmacology of existing
chemical matter with another potential target.
These 44 targets are identied for 137 target
combinations across 410,000 compounds
in this portfolio. Thus, by considering both
single-target and dual pharmacology approaches,
at least 181 potential prole opportunities can
be examined. As this is an analysis of known
chemical matter and biological activities,
many of these proles could be tested
immediately in appropriate disease models.
The network is represented in Cytoscape
117
.
Drug targets are color-coded by gene family: aminergic GPCRs, yellow; peptide GPCRs, orange; lipophilic GPCRs, light pink; ion channels, light blue; nuclear
hormone receptors, brown; phosphodiesterases, purple; protein kinases, pink; enzymes, green.
Pl3a
Pl3b
GR
IKKa
IKKb
SYK
P2X7
ITK
EP4
EGRF
LTB4
A1
A2B
A2A
M1
M3
M2
NK1
NK2 NK3
CCR3
CXCR2
5LO
2
1
3
1a
2c1
2a
2c
EP1
DP
EP2
DP1
CRTH2
H4
H1
H3
PDE7A
PDE4D
PDE7B
PDE4A
PDE4CPDE4B
RE VI E W
analysis of a corporate screening database of 70,000 biologically active
small molecules screened over 200 molecular assays, Hopkins et al.
observed a tight relationship between molecular weight and the
polypharmacology of a compound, in IC
50
measurements below
10 mM
25
. These empirical observations are complemented by theore-
tical work by Hann et al., in which using a simple model of ligand-
receptor pharmacophoric interaction revealed that the probability of a
randomly chosen ligand binding decreases precipitously as the ligand
becomes more complex
84
. Assuming that molecular weight is a proxy
measure of molecular complexity, these observations from experi-
ments and models provide the foundation for understanding the
success of fragment-based lead discovery
85
, in which a small library
(5002,000) of low-molecular-weight (100250 Da) compounds can
be successfully used to nd ligand-efcient
86
hits against a large
number of molecular targets. Therefore, it has been proposed that
multitarget fragment screens could be a promising approach for the
discovery of drug-like promiscuous ligands
87
. In particular, surface
plasmon resonance
88
may be well suited to multitarget fragment
screening owing to the increased sensitivity of the new generation of
instruments, the low protein consumption required compared with
NMR- and X-raybased fragment screening, and the multiple chan-
nels that can be screened simultaneously
89
.
Alternatively, advances in cheminformatics and the availability of
large-scale structure-activity relationship (SAR) databases provide the
tools necessary to develop computational methods of multitarget
design. Large-scale, integrated chemogenomics knowledge bases,
such as that described by Paolini et al., enable the systematic search
across large datasets of integrated structure-activity data for com-
pounds that are observed or predicted to bind to multiple targets
90
.
For example, 35% of biologically active compounds in the data
warehouse built by Paolini et al. are observed to bind to more than
one target, and the majority are active against
targets within the same gene family. However,
as we observed from the structure of a poly-
pharmacology interaction network that maps
12,000 interactions of 700 human proteins in
chemical space, there is a surprising level of
interaction between gene families (Fig. 1).
Though the predicted number of druggable
targets may be a relatively small fraction of
the proteome, the observed number of che-
mically tractable combinations (with integrated pharmacophores) is
over an order of magnitude larger
90
.
Wermuth expands this logic into a design strategy that can be used
for multitarget drug discovery called selective optimization of side
activities (SOSA)
91,92
. The SOSA idea provides a pragmatic approach
to designing polypharmacology: rather than attempting the difcult
problem of merging and integrating pharmacophores, the starting
point is an integrated pharmacophore that already provides some of
the nascent activity prole. An opportunistic strategy may be to
investigate, given the polypharmacology prole of a particular com-
pound, what phenotypic behavior it exhibits. For example, an analysis
of the polypharmacology interaction network between drug targets
associated with asthma is shown in Figure 2. An SOSA analysis of this
graph provides a wealth of opportunities for identifying lead series
that already exhibit interesting mixtures of pharmacology (for exam-
ple, the edges between the nodes) that were identied by integrating a
number of SAR data sources. As in all drug discovery, the choice of
lead matter is crucial in determining the success of a project.
Traditionally, medicinal chemists have attempted to combine activities
to create multitarget drugs. A more productive approach may be to
focus efforts on the discovery of lead compounds with interesting
multitarget proles from systematic data mining or multitarget
screening. Thus, one of the most pragmatic applications of network
pharmacology thinking could be to systematically reassess drug
candidates and drug discovery programs in the new light of under-
standing their wider biological activity proles
93
.
Multitarget design
The key to identifying multitarget drugs is in appreciating their
limits. Although the opportunity space for compounds with specic
polypharmacology proles may be signicantly larger than the
Figure 1 Human polypharmacology interactions
network at ten-fold selectivity. Network
representations of the integrated chemogenomics
space dened by Paolini et al.
90
. Two proteins
(nodes) are dened as interacting in chemical
space (edges) if they bind at least 10% of shared
screened compounds with a difference in potency
of only ten-fold below an activity cutoff of
10 mM (that is, if a compounds exhibits an
IC
50
10 nM against target A, it must show an
activity below IC
50
100 nM against target B to
be considered interacting in this network). At
these thresholds, 675 proteins are connected by
10,016 interactions in the total network by at
least one compound. Nodes are color-coded
by gene family: aminergic G proteincoupled
receptors (GPCRs), yellow; peptide GPCRs,
orange; other GPCRs, light pink; ion channels,
light blue; nuclear hormone receptors, brown;
phosphodiesterases, purple; protein kinases, pink;
enzymes, green; proteases, red; others, black.
RE VI E W
opportunity universe for single-target drugs (which is outlined by the
druggable genome
8
), not all target pair combinations will be accessible
to a single agent with drug-like properties. Predicting chemically
tractable combinations can increase the chances of nding a multi-
target compound. In order to develop a drug with a desired poly-
pharmacology prole, two problems need to be solved. First, a lead
compound needs to be identied with the desired biological activity
against multiple targets; then, this lead needs to be optimized into a
clinical candidate that combines the desired polypharmacology with a
safe, drug-like pharmaceutical prole.
The integration of in silico methods, combined with wide-ligand
biological proling against protein assays and gene expression arrays,
can provide drug designers with a new toolbox with which to assess
polypharmacology. First, proteins can be related by binding exact or
similar endogenous ligands or proteins; this can be determined by
exploiting data in gene ontologies and metabolic databases such as
KEGG
94,95
. Second, proteins can be related by known observed
polypharmacology of large sets of ligands such as those found in
pharmaceutical company screening sets and the medicinal chemistry
literature. A chemical network can be created that relates proteins that
bind the same ligand, where the strength of each edge can be
calculated using metrics of promiscuity
69,70,90,9699
.
If experimentally derived ligand data are not available, proteins can
be related by predictions of polypharmacology
90,100104
. Sets of ligands
for each protein obtained from chemogenomic data sources can be
used to train machine learning algorithms to predict pharmacology
activity proles. Bayesian approaches can be used to classify chemical
structures based on chemical ngerprints
90,100,101
. Dynamically
cross-comparing the chemical similarity of sets of ligands for each
protein is a complementary and promising method of predicting
polypharmacology proles from chemical structure
100,101
. The struc-
ture of the observed polypharmacology network itself also
provides information beyond chemical structure similarity that
could also be utilized using Bayesian network methods, as has
been the case in protein-protein interaction networks
105
. Though
promising, chemical ngerprintbased similarity methods paradoxi-
cally are incompatible with the two simplest relationshipsthose
of lipophilicity and molecular weight with promiscuityas larger
compounds tend to exhibit a greater number of ngerprints. The
development of improved polypharmacology prediction methods
will be an important topic in cheminformatics and toxicology.
Complementary to chemogenomics approaches to predict drug-
target relationships is the use of phenotypic data (such as clinical
side effects)
104,106
or gene expression proles
107
to cluster chemical
structures by functional effects. Clustering compounds by phenotypic
and gene signature proles enables unknown mechanisms to
be inferred.
Structural bioinformatics methods that map gene family sequence
alignments onto the structural information of protein binding sites
provide a valuable rst-order assessment of likely selectivity and
promiscuity within a gene family
95,108110
. Comparing protein binding
site similarity with the observed polypharmacology networks could
provide insight into protein binding site parameters that map with the
behavior of chemical networks. For example, analysis of the protein
kinase superfamily in the human genome reveals discrete clusters of
subfamilies when the full-length sequences are compared (Fig. 3).
However, these subfamily clusters break down when the sequence
similarity is measured at the level of the ATP binding site, where most
kinase inhibitors bind. Hence, at a rst approximation the difculties
drug designers have in ne tuning the selectivity proles of compe-
titive kinase inhibitors is not surprising
14,111,112
. Despite the inherent
challenge of achieving nely tuned kinase activity proles, deriving
kinase SAR data
112
from the wealth of large-scale chemogenomics
analyses and using subtle modication of existing kinase scaffolds to
exploit de-hydrons
111
are two promising strategies. Finally, if
three-dimensional protein structures of the targets of interest are
available, then parallel large-scale multitarget virtual screening is
also a promising method
113
.
Figure 2 Expanding opportunity for drug
discovery space with polypharmacology. A subset
of the network data shown in Figure 1 for
literature targets associated with asthma. Drug
targets are represented as nodes, and chemical
matter that binds to two or more nodes is
represented as edges. Targets are colored by gene
family. The color of the edges represents the
strength of the chemical network between two
targets as dened by the number of shared
compounds that are active against both targets
below an afnity of 1 mM: light blue (1 to
10 compounds) to black (41,000 compounds).
Of the 44 targets described in the literature
as potential drug targets for the treatment of
asthma, 44 share polypharmacology of existing
chemical matter with another potential target.
These 44 targets are identied for 137 target
combinations across 410,000 compounds
in this portfolio. Thus, by considering both
single-target and dual pharmacology approaches,
at least 181 potential prole opportunities can
be examined. As this is an analysis of known
chemical matter and biological activities,
many of these proles could be tested
immediately in appropriate disease models.
The network is represented in Cytoscape
117
.
Drug targets are color-coded by gene family: aminergic GPCRs, yellow; peptide GPCRs, orange; lipophilic GPCRs, light pink; ion channels, light blue; nuclear
hormone receptors, brown; phosphodiesterases, purple; protein kinases, pink; enzymes, green.
Pl3a
Pl3b
GR
IKKa
IKKb
SYK
P2X7
ITK
EP4
EGRF
LTB4
A1
A2B
A2A
M1
M3
M2
NK1
NK2 NK3
CCR3
CXCR2
5LO
2
1
3
1a
2c1
2a
2c
EP1
DP
EP2
DP1
CRTH2
H4
H1
H3
PDE7A
PDE4D
PDE7B
PDE4A
PDE4CPDE4B
RE VI E W
analysis of a corporate screening database of 70,000 biologically active
small molecules screened over 200 molecular assays, Hopkins et al.
observed a tight relationship between molecular weight and the
polypharmacology of a compound, in IC
50
measurements below
10 mM
25
. These empirical observations are complemented by theore-
tical work by Hann et al., in which using a simple model of ligand-
receptor pharmacophoric interaction revealed that the probability of a
randomly chosen ligand binding decreases precipitously as the ligand
becomes more complex
84
. Assuming that molecular weight is a proxy
measure of molecular complexity, these observations from experi-
ments and models provide the foundation for understanding the
success of fragment-based lead discovery
85
, in which a small library
(5002,000) of low-molecular-weight (100250 Da) compounds can
be successfully used to nd ligand-efcient
86
hits against a large
number of molecular targets. Therefore, it has been proposed that
multitarget fragment screens could be a promising approach for the
discovery of drug-like promiscuous ligands
87
. In particular, surface
plasmon resonance
88
may be well suited to multitarget fragment
screening owing to the increased sensitivity of the new generation of
instruments, the low protein consumption required compared with
NMR- and X-raybased fragment screening, and the multiple chan-
nels that can be screened simultaneously
89
.
Alternatively, advances in cheminformatics and the availability of
large-scale structure-activity relationship (SAR) databases provide the
tools necessary to develop computational methods of multitarget
design. Large-scale, integrated chemogenomics knowledge bases,
such as that described by Paolini et al., enable the systematic search
across large datasets of integrated structure-activity data for com-
pounds that are observed or predicted to bind to multiple targets
90
.
For example, 35% of biologically active compounds in the data
warehouse built by Paolini et al. are observed to bind to more than
one target, and the majority are active against
targets within the same gene family. However,
as we observed from the structure of a poly-
pharmacology interaction network that maps
12,000 interactions of 700 human proteins in
chemical space, there is a surprising level of
interaction between gene families (Fig. 1).
Though the predicted number of druggable
targets may be a relatively small fraction of
the proteome, the observed number of che-
mically tractable combinations (with integrated pharmacophores) is
over an order of magnitude larger
90
.
Wermuth expands this logic into a design strategy that can be used
for multitarget drug discovery called selective optimization of side
activities (SOSA)
91,92
. The SOSA idea provides a pragmatic approach
to designing polypharmacology: rather than attempting the difcult
problem of merging and integrating pharmacophores, the starting
point is an integrated pharmacophore that already provides some of
the nascent activity prole. An opportunistic strategy may be to
investigate, given the polypharmacology prole of a particular com-
pound, what phenotypic behavior it exhibits. For example, an analysis
of the polypharmacology interaction network between drug targets
associated with asthma is shown in Figure 2. An SOSA analysis of this
graph provides a wealth of opportunities for identifying lead series
that already exhibit interesting mixtures of pharmacology (for exam-
ple, the edges between the nodes) that were identied by integrating a
number of SAR data sources. As in all drug discovery, the choice of
lead matter is crucial in determining the success of a project.
Traditionally, medicinal chemists have attempted to combine activities
to create multitarget drugs. A more productive approach may be to
focus efforts on the discovery of lead compounds with interesting
multitarget proles from systematic data mining or multitarget
screening. Thus, one of the most pragmatic applications of network
pharmacology thinking could be to systematically reassess drug
candidates and drug discovery programs in the new light of under-
standing their wider biological activity proles
93
.
Multitarget design
The key to identifying multitarget drugs is in appreciating their
limits. Although the opportunity space for compounds with specic
polypharmacology proles may be signicantly larger than the
Figure 1 Human polypharmacology interactions
network at ten-fold selectivity. Network
representations of the integrated chemogenomics
space dened by Paolini et al.
90
. Two proteins
(nodes) are dened as interacting in chemical
space (edges) if they bind at least 10% of shared
screened compounds with a difference in potency
of only ten-fold below an activity cutoff of
10 mM (that is, if a compounds exhibits an
IC
50
10 nM against target A, it must show an
activity below IC
50
100 nM against target B to
be considered interacting in this network). At
these thresholds, 675 proteins are connected by
10,016 interactions in the total network by at
least one compound. Nodes are color-coded
by gene family: aminergic G proteincoupled
receptors (GPCRs), yellow; peptide GPCRs,
orange; other GPCRs, light pink; ion channels,
light blue; nuclear hormone receptors, brown;
phosphodiesterases, purple; protein kinases, pink;
enzymes, green; proteases, red; others, black.
RE VI E W
9. Austin, C.P. et al. The knockout mouse project. Nat. Genet. 36, 921924 (2004).
10. Zambrowicz, B.P., Turner, C.A. & Sands, A.T. Predicting drug efcacy: knockouts
model pipeline drugs of the pharmaceutical industry. Curr. Opin. Pharmacol. 3,
563570 (2003).
11. Zambrowicz, B.P. & Sands, A.T. Knockouts model the 100 best-selling drugswill
they model the next 100? Nat. Rev. Drug Discov. 2, 3851 (2003).
12. Kitano, H. Towards a theory of biological robustness. Mol. Syst. Biol. 3, 137 (2007).
13. Baraba si, A.L. & Oltvai, Z.N. Network biology: understanding the cells functional
organization. Nat. Rev. Genet. 5, 101113 (2004).
14. Albert, R., Jeong, H. & Barabasi, A.L. Error and attack tolerance of complex networks.
Nature 406, 378382 (2000).
15. Chen, Y. et al. Variations in DNA elucidate molecular networks that cause disease.
Nature 452, 429435 (2008).
16. Csermely, P., Agoston, V. & Pongor, S. The efciency of multi-target drugs: the
network approach might help drug design. Trends Pharmacol. Sci. 26, 178182
(2005).
17. Korcsma ros, T., Szalay, M.S., Bo de, C., Kova cs, I. & Csermely, P. How to design multi-
target drugs: target search options in cellular networks. Expert Opin. Drug Discov. 2,
110 (2007).
18. Ooi, S.L. et al. Global synthetic-lethality analysis and yeast functional proling.
Trends Genet. 22, 5663 (2006).
19. Hillenmeyer, M.E. et al. The chemical genomic portrait of yeast: uncovering a
phenotype for all genes. Science 320, 362365 (2008).
20. Roth, B.L., Shefer, D.J. & Kroeze, W.K. Magic shotguns versus magic bullets:
selectively non-selective drugs for mood disorders and schizophrenia. Nat. Rev. Drug
Discov. 3, 353359 (2004).
21. Wermuth, C.G. Multitarget drugs: the end of the one-target-one-disease philosophy?
Drug Discov. Today 9, 826827 (2004).
22. Keith, C.T., Borisy, A.A. & Stockwel, B.R. Multicomponent therapeutics for networked
systems. Nat. Rev. Drug Discov. 4, 7178 (2005).
23. Petrelli, A. & Giordano, S. From single- to multi-target drugs in cancer therapy: when
aspecicity becomes an advantage. Curr. Med. Chem. 15, 422432 (2008).
24. Mencher, S.K. & Wang, L.G. Promiscuous drugs compared to selective drugs
(promiscuity can be a virtue). BMC Clin. Pharmacol. 5, 3 (2005).
25. Hopkins, A.L., Mason, J.S. & Overington, J.P. Can we rationally design promiscuous
drugs? Curr. Opin. Struct. Biol. 16, 127136 (2006).
26. Flordellis, C.S., Manolis, A.S., Paris, H. & Karabinis, A. Rethinking target discovery in
polygenic diseases. Curr. Top. Med. Chem. 6, 17911798 (2006).
27. Dessalew, N. & Workalemahu, M. On the paradigm shift towards multitarget selective
drug design. Curr. Comput. Aided Drug Des. 4, 7690 (2008).
28. Keskin, O., Gursoy, A., Ma, B. & Nussinov, R. Towards drugs targeting multiple
proteins in a systems biology approach. Curr. Top. Med. Chem. 7, 943951
(2007).
29. Zimmermann, G.R., Leha r, J. & Keith, C.T. Multi-target therapeutics: when the whole
is greater than the sum of the parts. Drug Discov. Today 12, 3442 (2007).
30. McGovern, S.L., Caselli, E., Grigorieff, N. & Shoichet, B.K. A common mechanism
underlying promiscuous inhibitors from virtual and high-throughput screening.
J. Med. Chem. 45, 17121722 (2002).
31. Kerkela , R. et al. Cardiotoxicity of the cancer therapeutic agent imatinib mesylate.
Nat. Med. 12, 908916 (2006).
32. Kaelin, W.G., Jr. The concept of synthetic lethality in the context of anticancer
therapy. Nat. Rev. Cancer 5, 689698 (2005).
33. Gascoigne, K.E. & Taylor, S.S. Cancer cells display profound intra- and interline
variation following prolonged exposure to antimitotic drugs. Cancer Cell 14, 111122
(2008).
34. Whitehurst, A.W. et al. Synthetic lethal screen identication of chemosensitizer loci in
cancer cells. Nature 446, 815819 (2007).
35. Turner, N.C. et al. A synthetic lethal siRNA screen identifying genes mediating
sensitivity to a PARP inhibitor. EMBO J. 27, 13681377 (2008).
36. Gupta, G.P. et al. Mediators of vascular remodelling co-opted for sequential steps in
lung metastasis. Nature 446, 765770 (2007).
37. Eltarhouny, S.A. et al. Genes controlling spread of breast cancer to lung gang of 4.
Exp. Oncol. 30, 9195 (2008).
38. Kitano, H. A robustness-based approach to systems-oriented drug design. Nat. Rev.
Drug Discov. 6, 202210 (2007).
39. Bond, R.A. Can intellectualism stie scientic discovery? Nat. Rev. Drug Discov. 1,
825829 (2002).
40. Dudekula, N., Arora, V., Callaerts-Vegh, Z. & Bond, R.A. The temporal hormesis of
drug therapies. Dose Response 3, 414424 (2006).
41. Schwartz, G.K. & Shah, M.A. Targeting the cell cycle: a new approach to cancer
therapy. J. Clin. Oncol. 23, 94089421 (2005).
42. Mills, S.D. When will the genomics investment pay off for antibacterial discovery?
Biochem. Pharmacol. 71, 10961102 (2006).
43. Pucci, M.J. Use of genomics to select antibacterial targets. Biochem. Pharmacol. 71,
10661072 (2006).
44. Payne, D.J., Gwynn, M.N., Holmes, D.J. & Rosenberg, M. Genomic approaches to
antibacterial discovery. Methods Mol. Biol. 266, 231259 (2004).
45. Payne, D.J., Gwynn, M.N., Holmes, D.J. & Pompliano, D.L. Drugs for bad bugs:
confronting the challenges of antibacterial discovery. Nat. Rev. Drug Discov. 6, 2940
(2007).
46. Lange, R.P., Locher, H.H., Wyss, P.C. & Then, R.L. The targets of currently used
antibacterial agents: lessons for drug discovery. Curr. Pharm. Des. 13, 31403154
(2007).
47. Denome, S.A., Elf, P.K., Henderson, T.A., Nelson, D.E. & Young, K.D. Escherichia coli
mutants lacking all possible combinations of eight penicillin binding proteins:
viability, characteristics, and implications for peptidoglycan synthesis. J. Bacteriol.
181, 39813993 (1999).
48. Janoir, C., Zeller, V., Kitzis, M.D., Moreau, N.J. & Gutmann, L. High-level uoroqui-
nolone resistance in Streptococcus pneumoniae requires mutations in parC and gyrA.
Antimicrob. Agents Chemother. 40, 27602764 (1996).
49. Ramaswamy, S. Rational design of cancer-drug combinations. N. Engl. J. Med. 357,
299300 (2007).
50. Mayer, L.D. & Janoff, A.S. Optimizing combination chemotherapy by controlling drug
ratios. Mol. Interv. 7, 216223 (2007).
51. Dancey, J.E. & Chen, H.X. Strategies for optimizing combinations of molecularly
targeted anticancer agents. Nat. Rev. Drug Discov. 5, 649659 (2006).
52. Borisy, A.A. et al. Systematic discovery of multicomponent therapeutics. Proc. Natl.
Acad. Sci. USA 100, 79777982 (2003).
53. Radhakrishnan, M.L. & Tidor, B. Optimal drug cocktail design: methods for targeting
molecular ensembles and insights from theoretical model systems. J. Chem. Inf.
Model. 48, 10551073 (2008).
54. Tsui, I.F., Chari, R., Buys, T.P. & Lam, W.L. Public databases and software for the
pathway analysis of cancer genomes. Cancer Inform. 3, 389407 (2007).
55. Swanson, D.R. Medical literature as a potential source of new knowledge. Bull. Med.
Libr. Assoc. 78, 2937 (1990).
56. Wren, J.D., Bekeredjian, R., Stewart, J.A., Shohet, R.V. & Garner, H.R. Knowledge
discovery by automated identication and ranking of implicit relationships. Bioinfor-
matics 20, 389398 (2004).
57. Bekhuis, T. Conceptual biology, hypothesis discovery, and text mining: Swansons
legacy. Biomed. Digit. Libr. 3, 2 (2006).
58. Loging, W., Harland, L. & Williams-Jones, B. High-throughput electronic biology:
mining information for drug discovery. Nat. Rev. Drug Discov. 6, 220230 (2007).
59. Peirce, C.S. The Essential Peirce Vol. 2 (Indiana University Press, Bloomington,
Indiana, USA, 1998).
60. Kell, D.B. & Oliver, S.G. Here is the evidence, now what is the hypothesis? The
complementary roles of inductive and hypothesis-driven science in the post-genomic
era. Bioessays 26, 99105 (2004).
61. Latora, V. & Marchiori, M. Vulnerability and protection of infrastructure networks.
Phys. Rev. E 71, 015103R (2005).
62. Jeong, H., Mason, S.P., Baraba si, A.L. & Oltvai, Z.N. Lethality and centrality in
protein networks. Nature 411, 4142 (2001).
63. Coulomb, S., Bauer, M., Bernard, D. & Marsolier-Kergoat, M.C. Gene essentiality and
the topology of protein interaction networks. Proc. Biol. Sci. 272, 17211725
(2005).
64. Joy, M.P., Brock, A., Ingber, D.E. & Huang, S. High-betweenness proteins in the yeast
protein interaction network. J. Biomed. Biotechnol. 2, 96103 (2005).
65. Yu, H., Kim, P.M., Sprecher, E., Trifonov, V. & Gerstein, M. The importance of
bottlenecks in protein networks: correlation with gene essentiality and expression
dynamics. PLoS Comput. Biol. 3, e59 (2007).
66. Hwang, W.C., Zhang, A. & Ramanathan, M. Identication of information ow-
modulating drug targets: a novel bridging paradigm for drug discovery. Clin. Pharma-
col. Ther. published online, doi:10.1038/clpt.2008.129 (9 July 2008).
67. Hahn, M.W. & Kern, A.D. Comparative genomics of centrality and essentiality in
three eukaryotic protein-interaction networks. Mol. Biol. Evol. 22, 803806
(2005).
68. Han, J.D. et al. Evidence for dynamically organized modularity in the yeast protein-
protein interaction network. Nature 430, 8893 (2004).
69. Yildirim, M.A., Goh, K.I., Cusick, M.E., Baraba si, A.L. & Vidal, M. Drug-target
network. Nat. Biotechnol. 25, 11191126 (2007).
70. Nacher, J.C. & Schwartz, J.M. A global view of drug-therapy interactions. BMC
Pharmacol. 8, 5 (2008).
71. Motter, A.E., Nishikawa, T. & Lai, Y. Range-based attack on links in scale-free
networks: are long-range links responsible for the small-world phenomenon? Phys.
Rev. E 66, 065103 (2002).
72. Moriya, H., Shimizu-Yoshida, Y. & Kitano, H. In vivo robustness analysis of
cell division cycle genes in Saccharomyces cerevisiae. PLoS Genet. 2, e111
(2006).
73. Wunderlich, Z. & Mirny, L.A. Using the topology of metabolic networks to predict
viability of mutant strains. Biophys. J. 91, 23042311 (2006).
74. Kovacs, I., Csermely, P., Korcsmaros, T. & Szalay, M. WO patent application WO
2007-IB50471 2007 0213 (2007).
75. Watterson, S., Marshall, S. & Ghazal, P. Logic models of pathway biology. Drug Discov.
Today 13, 447456 (2008).
76. Gerber, S., Assmus, H., Bakker, B. & Klipp, E. Drug-efcacy depends on the inhibitor
type and the target position in a metabolic networka systematic study. J. Theor.
Biol. 252, 442455 (2008).
77. Potapov, A.P., Goemann, B. & Wingender, E. The pairwise disconnectivity index as a
new metric for the topological analysis of regulatory networks. BMC Bioinformatics 9,
227 (2008).
78. Agoston, V., Csermely, P. & Pongor, S. Multiple weak hits confuse complex systems: a
transcriptional regulatory network as an example. Phys. Rev. E 71, 051909
(2005).
79. Motter, A.E., Gulbahce, N., Almaas, E. & Baraba si, A.L. Predicting synthetic rescues
in metabolic networks. Mol. Syst. Biol. 4, 168 (2008).
80. Leeson, P.D. & Springthorpe, B. The inuence of drug-like concepts on decision-
making in medicinal chemistry. Nat. Rev. Drug Discov. 6, 881890 (2007).
RE VI E W
Fusing multiple prediction methods is likely to improve the overall
success rates of in silico lead identication. Once a lead is identied,
the second computational design problem lies in the constraints of
optimizing in multiple dimensions. If we assume, as the data suggest,
that very few drugs are truly selective, then most biologically active
small molecules have a degree of promiscuity by their nature. The
challenge for medicinal chemists is to understand the prole of each
compound, to ne tune the prole, and then to select for clinical
development those compounds whose proles maximally modulate a
disease network with the minimum level of toxicity and side effects.
Conclusion
Network pharmacology is an approach to drug design that encom-
passes systems biology, network analysis, connectivity, redundancy
and pleiotropy. Network pharmacology offers a way of thinking about
drug discovery that simultaneously embraces efforts to improve
clinical efcacy and understand side effects and toxicitytwo of the
most important reasons for failure. Avariety of studies have shown the
power of network analysis in understanding biological systems.
Furthermore, emergent phenotypes beyond those seen in single-gene
deletion experiments have been observed through synthetic behaviors,
combinations and chemical biology probes. The biological rationale
for considering multitarget strategies over single-target approaches is
compelling, yet such strategies are at present a minority activity in the
pharmaceutical industry. The reason is that optimizing multiple
activities, while trying to balance drug-like properties and control
unwanted off-target effects, is a difcult task. We do not yet have a
robust set of design tools with which to apply this approach routinely.
Structure-based drug design took nearly two decades of multiple,
parallel technological improvements to arrive at its current main-
stream position in medicinal chemistry. Developments in computer
graphics, high-power radiation sources, computational processing
power, renement protocols, virtual screening and cryocrystallography
were all necessary to create the environment for rapid, iterative
structure-based drug discovery. To make network pharmacology
commonplace, a different set of tools, concerned with combinatorial
and network search algorithms and methods for predicting the
biological proles, will need to be rened. Network pharmacology
re-introduces the old idea that understanding the biological and
kinetic prole of the drug is more important than individual valida-
tion of targets or combinations of targets. In many ways, the network
pharmacology strategy outlined here is a modern reinvention of Paul
Janssens original and incredibly successful methods of drug discovery,
where side activities of compounds are explored through a broad
spectrum of structure-activity relationships
114
. Given the crisis in
translation facing the pharmaceutical industry, network pharmacology
offers a new framework for thinking about how to innovate drug
discovery, and thus it is an idea whose time has come.
Published online at http://www.nature.com/naturechemicalbiology/
Reprints and permissions information is available online at http://npg.nature.com/
reprintsandpermissions/
1. Kola, I. & Landis, J. Can the pharmaceutical industry reduce attrition rates? Nat. Rev.
Drug Discov. 3, 711716 (2004).
2. Sams-Dodd, F. Target-based drug discovery: is something wrong? Drug Discov. Today
10, 139147 (2005).
3. Kaufmann, S.H.E. Paul Ehrlich: founder of chemotherapy. Nat. Rev. Drug Discov. 7,
373 (2008).
4. Zambrowicz, B.P. & Sands, A.T. Modeling drug action in the mouse with knockouts
and RNA interference. Drug Discov. Today Targets 3, 198207 (2004).
5. Winzeler, E.A. et al. Functional characterization of the S. cerevisiae genome by gene
deletion and parallel analysis. Science 285, 901906 (1999).
6. Giaever, G. et al. Functional proling of the Saccharomyces cerevisiae genome.
Nature 418, 387391 (2002).
7. Deutschbauer, A.M. et al. Mechanisms of haploinsufciency revealed by genome-wide
proling in yeast. Genetics 169, 19151925 (2005).
8. Hopkins, A.L. & Groom, C.R. The druggable genome. Nat. Rev. Drug Discov. 1,
727730 (2002).
>gi|1170188|sp|P08631|HCK_HUMAN TYROSINE-PROTEIN KINASE HCK
MGGRSSCEDPGCPRDEERAPRMGSMKSKFLQVGGNTFSKTETSASPHCPVYVPDPTSTIKPGPNSHNSNTPGIREAGSEDIIVVALYDYEAIHHEDL
SFQKGDQMVVLEESGEWWKARSLATRKEGYIPSNYVARVDSLETEEWFFKGISRKDAERQLLAPGNMLGSFMIRDSETTKGSYSLSVRDYDPRQGDT
VKHYKIRTLDNGGFYISPRSTFSTLQELVDHYKKGNDGLCQKLSVPCMSSKPQKPWEKDAWEIPRESLKLEKKLGAGQFGEVWMATYNKHTKVAVKT
MKPGSMSVEAFLAEANVMKTLQHDKLVKLHAVVTKEPIYIITEFMAKGSLLDFLKSDEGSKQPLPKLIDFSAQIAEGMAFIEQRNYIHRDLRAANIL
VSASLVCKIADFGLARVIEDNEYTAREGAKFPIKWTAPEAINFGSFTIKSDVWSFGILLMEIVTYGRIPYPGMSNPEVIRALERGYRMPRPENCPEE
LYNIMMRCWKNRPEERPTFEYIQSVLDDFYTATESQYQQQP
a
b c d
Figure 3 Protein kinase inhibitor promiscuity as a function of binding site sequence similarity. (a,b) The full-length sequence of human protein tyrosine
kinase HCK (a), where the amino acids surrounding the ATP binding site are color-coded by their distance from the binding site surface when mapped onto
the canonical protein kinase structure (b). (c) Multidimensional scaling of the human kinome to cluster kinases using full sequences reveals that the kinases
cluster into discrete families. (d) Multidimensional scaling of the same kinases using the binding siteweighted sequences as dened in a and b reveals the
breakdown of the subfamily clustering and the similarity at the binding site level of many diverse protein kinases. Graphic courtesy of Colin Groom
(Cambridge Crystallographic Data Centre, Cambridge, UK).
RE VI E W
9. Austin, C.P. et al. The knockout mouse project. Nat. Genet. 36, 921924 (2004).
10. Zambrowicz, B.P., Turner, C.A. & Sands, A.T. Predicting drug efcacy: knockouts
model pipeline drugs of the pharmaceutical industry. Curr. Opin. Pharmacol. 3,
563570 (2003).
11. Zambrowicz, B.P. & Sands, A.T. Knockouts model the 100 best-selling drugswill
they model the next 100? Nat. Rev. Drug Discov. 2, 3851 (2003).
12. Kitano, H. Towards a theory of biological robustness. Mol. Syst. Biol. 3, 137 (2007).
13. Baraba si, A.L. & Oltvai, Z.N. Network biology: understanding the cells functional
organization. Nat. Rev. Genet. 5, 101113 (2004).
14. Albert, R., Jeong, H. & Barabasi, A.L. Error and attack tolerance of complex networks.
Nature 406, 378382 (2000).
15. Chen, Y. et al. Variations in DNA elucidate molecular networks that cause disease.
Nature 452, 429435 (2008).
16. Csermely, P., Agoston, V. & Pongor, S. The efciency of multi-target drugs: the
network approach might help drug design. Trends Pharmacol. Sci. 26, 178182
(2005).
17. Korcsma ros, T., Szalay, M.S., Bo de, C., Kova cs, I. & Csermely, P. How to design multi-
target drugs: target search options in cellular networks. Expert Opin. Drug Discov. 2,
110 (2007).
18. Ooi, S.L. et al. Global synthetic-lethality analysis and yeast functional proling.
Trends Genet. 22, 5663 (2006).
19. Hillenmeyer, M.E. et al. The chemical genomic portrait of yeast: uncovering a
phenotype for all genes. Science 320, 362365 (2008).
20. Roth, B.L., Shefer, D.J. & Kroeze, W.K. Magic shotguns versus magic bullets:
selectively non-selective drugs for mood disorders and schizophrenia. Nat. Rev. Drug
Discov. 3, 353359 (2004).
21. Wermuth, C.G. Multitarget drugs: the end of the one-target-one-disease philosophy?
22. Keith, C.T., Borisy, A.A. & Stockwel, B.R. Multicomponent therapeutics for networked
systems. Nat. Rev. Drug Discov. 4, 7178 (2005).
23. Petrelli, A. & Giordano, S. From single- to multi-target drugs in cancer therapy: when
aspecicity becomes an advantage. Curr. Med. Chem. 15, 422432 (2008).
24. Mencher, S.K. & Wang, L.G. Promiscuous drugs compared to selective drugs
(promiscuity can be a virtue). BMC Clin. Pharmacol. 5, 3 (2005).
25. Hopkins, A.L., Mason, J.S. & Overington, J.P. Can we rationally design promiscuous
drugs? Curr. Opin. Struct. Biol. 16, 127136 (2006).
26. Flordellis, C.S., Manolis, A.S., Paris, H. & Karabinis, A. Rethinking target discovery in
polygenic diseases. Curr. Top. Med. Chem. 6, 17911798 (2006).
27. Dessalew, N. & Workalemahu, M. On the paradigm shift towards multitarget selective
drug design. Curr. Comput. Aided Drug Des. 4, 7690 (2008).
28. Keskin, O., Gursoy, A., Ma, B. & Nussinov, R. Towards drugs targeting multiple
proteins in a systems biology approach. Curr. Top. Med. Chem. 7, 943951
(2007).
29. Zimmermann, G.R., Leha r, J. & Keith, C.T. Multi-target therapeutics: when the whole
is greater than the sum of the parts. Drug Discov. Today 12, 3442 (2007).
30. McGovern, S.L., Caselli, E., Grigorieff, N. & Shoichet, B.K. A common mechanism
underlying promiscuous inhibitors from virtual and high-throughput screening.
J. Med. Chem. 45, 17121722 (2002).
31. Kerkela , R. et al. Cardiotoxicity of the cancer therapeutic agent imatinib mesylate.
Nat. Med. 12, 908916 (2006).
32. Kaelin, W.G., Jr. The concept of synthetic lethality in the context of anticancer
therapy. Nat. Rev. Cancer 5, 689698 (2005).
33. Gascoigne, K.E. & Taylor, S.S. Cancer cells display profound intra- and interline
variation following prolonged exposure to antimitotic drugs. Cancer Cell 14, 111122
(2008).
34. Whitehurst, A.W. et al. Synthetic lethal screen identication of chemosensitizer loci in
cancer cells. Nature 446, 815819 (2007).
35. Turner, N.C. et al. A synthetic lethal siRNA screen identifying genes mediating
sensitivity to a PARP inhibitor. EMBO J. 27, 13681377 (2008).
36. Gupta, G.P. et al. Mediators of vascular remodelling co-opted for sequential steps in
lung metastasis. Nature 446, 765770 (2007).
37. Eltarhouny, S.A. et al. Genes controlling spread of breast cancer to lung gang of 4.
Exp. Oncol. 30, 9195 (2008).
38. Kitano, H. A robustness-based approach to systems-oriented drug design. Nat. Rev.
Drug Discov. 6, 202210 (2007).
39. Bond, R.A. Can intellectualism stie scientic discovery? Nat. Rev. Drug Discov. 1,
825829 (2002).
40. Dudekula, N., Arora, V., Callaerts-Vegh, Z. & Bond, R.A. The temporal hormesis of
drug therapies. Dose Response 3, 414424 (2006).
41. Schwartz, G.K. & Shah, M.A. Targeting the cell cycle: a new approach to cancer
therapy. J. Clin. Oncol. 23, 94089421 (2005).
42. Mills, S.D. When will the genomics investment pay off for antibacterial discovery?
Biochem. Pharmacol. 71, 10961102 (2006).
43. Pucci, M.J. Use of genomics to select antibacterial targets. Biochem. Pharmacol. 71,
10661072 (2006).
44. Payne, D.J., Gwynn, M.N., Holmes, D.J. & Rosenberg, M. Genomic approaches to
antibacterial discovery. Methods Mol. Biol. 266, 231259 (2004).
45. Payne, D.J., Gwynn, M.N., Holmes, D.J. & Pompliano, D.L. Drugs for bad bugs:
confronting the challenges of antibacterial discovery. Nat. Rev. Drug Discov. 6, 2940
(2007).
46. Lange, R.P., Locher, H.H., Wyss, P.C. & Then, R.L. The targets of currently used
antibacterial agents: lessons for drug discovery. Curr. Pharm. Des. 13, 31403154
(2007).
47. Denome, S.A., Elf, P.K., Henderson, T.A., Nelson, D.E. & Young, K.D. Escherichia coli
mutants lacking all possible combinations of eight penicillin binding proteins:
viability, characteristics, and implications for peptidoglycan synthesis. J. Bacteriol.
181, 39813993 (1999).
48. Janoir, C., Zeller, V., Kitzis, M.D., Moreau, N.J. & Gutmann, L. High-level uoroqui-
nolone resistance in Streptococcus pneumoniae requires mutations in parC and gyrA.
Antimicrob. Agents Chemother. 40, 27602764 (1996).
49. Ramaswamy, S. Rational design of cancer-drug combinations. N. Engl. J. Med. 357,
299300 (2007).
50. Mayer, L.D. & Janoff, A.S. Optimizing combination chemotherapy by controlling drug
ratios. Mol. Interv. 7, 216223 (2007).
51. Dancey, J.E. & Chen, H.X. Strategies for optimizing combinations of molecularly
targeted anticancer agents. Nat. Rev. Drug Discov. 5, 649659 (2006).
52. Borisy, A.A. et al. Systematic discovery of multicomponent therapeutics. Proc. Natl.
Acad. Sci. USA 100, 79777982 (2003).
53. Radhakrishnan, M.L. & Tidor, B. Optimal drug cocktail design: methods for targeting
molecular ensembles and insights from theoretical model systems. J. Chem. Inf.
Model. 48, 10551073 (2008).
54. Tsui, I.F., Chari, R., Buys, T.P. & Lam, W.L. Public databases and software for the
pathway analysis of cancer genomes. Cancer Inform. 3, 389407 (2007).
55. Swanson, D.R. Medical literature as a potential source of new knowledge. Bull. Med.
Libr. Assoc. 78, 2937 (1990).
56. Wren, J.D., Bekeredjian, R., Stewart, J.A., Shohet, R.V. & Garner, H.R. Knowledge
discovery by automated identication and ranking of implicit relationships. Bioinfor-
matics 20, 389398 (2004).
57. Bekhuis, T. Conceptual biology, hypothesis discovery, and text mining: Swansons
legacy. Biomed. Digit. Libr. 3, 2 (2006).
58. Loging, W., Harland, L. & Williams-Jones, B. High-throughput electronic biology:
mining information for drug discovery. Nat. Rev. Drug Discov. 6, 220230 (2007).
59. Peirce, C.S. The Essential Peirce Vol. 2 (Indiana University Press, Bloomington,
Indiana, USA, 1998).
60. Kell, D.B. & Oliver, S.G. Here is the evidence, now what is the hypothesis? The
complementary roles of inductive and hypothesis-driven science in the post-genomic
era. Bioessays 26, 99105 (2004).
61. Latora, V. & Marchiori, M. Vulnerability and protection of infrastructure networks.
Phys. Rev. E 71, 015103R (2005).
62. Jeong, H., Mason, S.P., Baraba si, A.L. & Oltvai, Z.N. Lethality and centrality in
protein networks. Nature 411, 4142 (2001).
63. Coulomb, S., Bauer, M., Bernard, D. & Marsolier-Kergoat, M.C. Gene essentiality and
the topology of protein interaction networks. Proc. Biol. Sci. 272, 17211725
(2005).
64. Joy, M.P., Brock, A., Ingber, D.E. & Huang, S. High-betweenness proteins in the yeast
protein interaction network. J. Biomed. Biotechnol. 2, 96103 (2005).
65. Yu, H., Kim, P.M., Sprecher, E., Trifonov, V. & Gerstein, M. The importance of
bottlenecks in protein networks: correlation with gene essentiality and expression
dynamics. PLoS Comput. Biol. 3, e59 (2007).
66. Hwang, W.C., Zhang, A. & Ramanathan, M. Identication of information ow-
modulating drug targets: a novel bridging paradigm for drug discovery. Clin. Pharma-
col. Ther. published online, doi:10.1038/clpt.2008.129 (9 July 2008).
67. Hahn, M.W. & Kern, A.D. Comparative genomics of centrality and essentiality in
three eukaryotic protein-interaction networks. Mol. Biol. Evol. 22, 803806
(2005).
68. Han, J.D. et al. Evidence for dynamically organized modularity in the yeast protein-
protein interaction network. Nature 430, 8893 (2004).
69. Yildirim, M.A., Goh, K.I., Cusick, M.E., Baraba si, A.L. & Vidal, M. Drug-target
network. Nat. Biotechnol. 25, 11191126 (2007).
70. Nacher, J.C. & Schwartz, J.M. A global view of drug-therapy interactions. BMC
Pharmacol. 8, 5 (2008).
71. Motter, A.E., Nishikawa, T. & Lai, Y. Range-based attack on links in scale-free
networks: are long-range links responsible for the small-world phenomenon? Phys.
Rev. E 66, 065103 (2002).
72. Moriya, H., Shimizu-Yoshida, Y. & Kitano, H. In vivo robustness analysis of
cell division cycle genes in Saccharomyces cerevisiae. PLoS Genet. 2, e111
(2006).
73. Wunderlich, Z. & Mirny, L.A. Using the topology of metabolic networks to predict
viability of mutant strains. Biophys. J. 91, 23042311 (2006).
74. Kovacs, I., Csermely, P., Korcsmaros, T. & Szalay, M. WO patent application WO
2007-IB50471 2007 0213 (2007).
75. Watterson, S., Marshall, S. & Ghazal, P. Logic models of pathway biology. Drug Discov.
Today 13, 447456 (2008).
76. Gerber, S., Assmus, H., Bakker, B. & Klipp, E. Drug-efcacy depends on the inhibitor
type and the target position in a metabolic networka systematic study. J. Theor.
Biol. 252, 442455 (2008).
77. Potapov, A.P., Goemann, B. & Wingender, E. The pairwise disconnectivity index as a
new metric for the topological analysis of regulatory networks. BMC Bioinformatics 9,
227 (2008).
78. Agoston, V., Csermely, P. & Pongor, S. Multiple weak hits confuse complex systems: a
transcriptional regulatory network as an example. Phys. Rev. E 71, 051909
(2005).
79. Motter, A.E., Gulbahce, N., Almaas, E. & Baraba si, A.L. Predicting synthetic rescues
in metabolic networks. Mol. Syst. Biol. 4, 168 (2008).
80. Leeson, P.D. & Springthorpe, B. The inuence of drug-like concepts on decision-
making in medicinal chemistry. Nat. Rev. Drug Discov. 6, 881890 (2007).
RE VI E W
Fusing multiple prediction methods is likely to improve the overall
success rates of in silico lead identication. Once a lead is identied,
the second computational design problem lies in the constraints of
optimizing in multiple dimensions. If we assume, as the data suggest,
that very few drugs are truly selective, then most biologically active
small molecules have a degree of promiscuity by their nature. The
challenge for medicinal chemists is to understand the prole of each
compound, to ne tune the prole, and then to select for clinical
development those compounds whose proles maximally modulate a
disease network with the minimum level of toxicity and side effects.
Conclusion
Network pharmacology is an approach to drug design that encom-
passes systems biology, network analysis, connectivity, redundancy
and pleiotropy. Network pharmacology offers a way of thinking about
drug discovery that simultaneously embraces efforts to improve
clinical efcacy and understand side effects and toxicitytwo of the
most important reasons for failure. Avariety of studies have shown the
power of network analysis in understanding biological systems.
Furthermore, emergent phenotypes beyond those seen in single-gene
deletion experiments have been observed through synthetic behaviors,
combinations and chemical biology probes. The biological rationale
for considering multitarget strategies over single-target approaches is
compelling, yet such strategies are at present a minority activity in the
pharmaceutical industry. The reason is that optimizing multiple
activities, while trying to balance drug-like properties and control
unwanted off-target effects, is a difcult task. We do not yet have a
robust set of design tools with which to apply this approach routinely.
Structure-based drug design took nearly two decades of multiple,
parallel technological improvements to arrive at its current main-
stream position in medicinal chemistry. Developments in computer
graphics, high-power radiation sources, computational processing
power, renement protocols, virtual screening and cryocrystallography
were all necessary to create the environment for rapid, iterative
structure-based drug discovery. To make network pharmacology
commonplace, a different set of tools, concerned with combinatorial
and network search algorithms and methods for predicting the
biological proles, will need to be rened. Network pharmacology
re-introduces the old idea that understanding the biological and
kinetic prole of the drug is more important than individual valida-
tion of targets or combinations of targets. In many ways, the network
pharmacology strategy outlined here is a modern reinvention of Paul
Janssens original and incredibly successful methods of drug discovery,
where side activities of compounds are explored through a broad
spectrum of structure-activity relationships
114
. Given the crisis in
translation facing the pharmaceutical industry, network pharmacology
offers a new framework for thinking about how to innovate drug
discovery, and thus it is an idea whose time has come.
Published online at http://www.nature.com/naturechemicalbiology/
Reprints and permissions information is available online at http://npg.nature.com/
reprintsandpermissions/
1. Kola, I. & Landis, J. Can the pharmaceutical industry reduce attrition rates? Nat. Rev.
Drug Discov. 3, 711716 (2004).
2. Sams-Dodd, F. Target-based drug discovery: is something wrong? Drug Discov. Today
10, 139147 (2005).
3. Kaufmann, S.H.E. Paul Ehrlich: founder of chemotherapy. Nat. Rev. Drug Discov. 7,
373 (2008).
4. Zambrowicz, B.P. & Sands, A.T. Modeling drug action in the mouse with knockouts
and RNA interference. Drug Discov. Today Targets 3, 198207 (2004).
5. Winzeler, E.A. et al. Functional characterization of the S. cerevisiae genome by gene
deletion and parallel analysis. Science 285, 901906 (1999).
6. Giaever, G. et al. Functional proling of the Saccharomyces cerevisiae genome.
Nature 418, 387391 (2002).
7. Deutschbauer, A.M. et al. Mechanisms of haploinsufciency revealed by genome-wide
proling in yeast. Genetics 169, 19151925 (2005).
8. Hopkins, A.L. & Groom, C.R. The druggable genome. Nat. Rev. Drug Discov. 1,
727730 (2002).
>gi|1170188|sp|P08631|HCK_HUMAN TYROSINE-PROTEIN KINASE HCK
MGGRSSCEDPGCPRDEERAPRMGSMKSKFLQVGGNTFSKTETSASPHCPVYVPDPTSTIKPGPNSHNSNTPGIREAGSEDIIVVALYDYEAIHHEDL
SFQKGDQMVVLEESGEWWKARSLATRKEGYIPSNYVARVDSLETEEWFFKGISRKDAERQLLAPGNMLGSFMIRDSETTKGSYSLSVRDYDPRQGDT
VKHYKIRTLDNGGFYISPRSTFSTLQELVDHYKKGNDGLCQKLSVPCMSSKPQKPWEKDAWEIPRESLKLEKKLGAGQFGEVWMATYNKHTKVAVKT
MKPGSMSVEAFLAEANVMKTLQHDKLVKLHAVVTKEPIYIITEFMAKGSLLDFLKSDEGSKQPLPKLIDFSAQIAEGMAFIEQRNYIHRDLRAANIL
VSASLVCKIADFGLARVIEDNEYTAREGAKFPIKWTAPEAINFGSFTIKSDVWSFGILLMEIVTYGRIPYPGMSNPEVIRALERGYRMPRPENCPEE
LYNIMMRCWKNRPEERPTFEYIQSVLDDFYTATESQYQQQP
a
b c d
Figure 3 Protein kinase inhibitor promiscuity as a function of binding site sequence similarity. (a,b) The full-length sequence of human protein tyrosine
kinase HCK (a), where the amino acids surrounding the ATP binding site are color-coded by their distance from the binding site surface when mapped onto
the canonical protein kinase structure (b). (c) Multidimensional scaling of the human kinome to cluster kinases using full sequences reveals that the kinases
cluster into discrete families. (d) Multidimensional scaling of the same kinases using the binding siteweighted sequences as dened in a and b reveals the
breakdown of the subfamily clustering and the similarity at the binding site level of many diverse protein kinases. Graphic courtesy of Colin Groom
(Cambridge Crystallographic Data Centre, Cambridge, UK).
RE VI E W
Investment in drug research and development (R&D)
has increased substantially in recent decades, but the
annual number of truly innovative new medicines
approved by the US Food and Drug Administration
(FDA) has not increased accordingly, and attrition rates
are very high
1
. Indeed, in a recent analysis
2
it was noted
that without a dramatic improvement in R&D produc-
tivity, the pharmaceutical industry cannot sustain suf-
ficient innovation to replace the loss of revenues due to
patent expirations for successful products.
The authors of this analysis
2
also considered R&D pro-
ductivity in two dimensions: efficiency and effectiveness.
R&D efficiency represents the ability to translate inputs
(such as ideas, investment and effort) into defined out-
puts (such as milestones that represent resolved uncer-
tainties), whereas R&D effectiveness can be considered as
the ability to produce outputs with certain intended and
desired qualities. A key efficiency variable for increased
productivity is the probability of technical success. If the
probability of technical success could be increased (by
reducing attrition) for any given drug candidate or, ide-
ally, for a portfolio of drug candidates, then productivity
would increase accordingly. The authors also suggested
that target selection may be one of the most important
determinants of attrition and overall R&D productivity
2
.
Since the dawn of the genomics era in the 1990s, the
main focus of drug discovery has been on drug targets,
which are typically proteins that appear to have a key
role in disease pathogenesis
35
. Modification of target
activity provides a rational basis for the discovery of
new medicines; a target-centric approach provides a
specific biological hypothesis to be tested and a starting
point for the identification of molecules to do this with.
Tremendous advances have been made in the develop-
ment of new tools to identify targets (for example, RNA
interference) and compounds that interact with these
targets (for example, high-throughput target-based
screening assays that are applicable to key protein fami-
lies such as G protein-coupled receptors and kinases).
Structure-based tools that can be used to aid lead
identification and optimization for some targets have
also been developed, including X-ray crystallography
and computational modelling and screening (virtual
screening).
However, despite the power of these tools to identify
potential drug candidates, R&D productivity remains a
crucial challenge for the pharmaceutical industry, which
raises questions about the possible limitations of a tar-
get-centric approach to drug discovery. Indeed, before
the introduction of target-based approaches, drug dis-
covery was driven primarily by phenotypic assays, often
with limited knowledge of the molecular mechanisms
of disease. Nevertheless, the pharmaceutical industry
was successful in the discovery and development of new
*Roche Palo Alto,
3431 Hillview Avenue,
Palo Alto, California 94304,
USA.
iRND3 (Institute for Rare and

Neglected Diseases Drug
Discovery), 951 Old County
Road, PMB 316, Belmont,
California 940022760, USA.
Correspondence to D.C.S.
email david.swinney@irnd3.
org
doi:10.1038/nrd3480
How were new medicines discovered?
David C.Swinney*
and Jason Anthony*

Abstract | Preclinical strategies that are used to identify potential drug candidates include
target-based screening, phenotypic screening, modification of natural substances and
biologic-based approaches. To investigate whether some strategies have been more
successful than others in the discovery of new drugs, we analysed the discovery strategies
and the molecular mechanism of action (MMOA) for new molecular entities and new
biologics that were approved by the US Food and Drug Administration between 1999 and
2008. Out of the 259 agents that were approved, 75 were first-in-class drugs with new
MMOAs, and out of these, 50 (67%) were small molecules and 25 (33%) were biologics.
The results also show that the contribution of phenotypic screening to the discovery of
first-in-class small-molecule drugs exceeded that of target-based approaches with 28
and 17 of these drugs coming from the two approaches, respectively in an era in which
the major focus was on target-based approaches. We postulate that a target-centric
approach for first-in-class drugs, without consideration of an optimal MMOA, may
contribute to the current high attrition rates and low productivity in pharmaceutical
research and development.
NATURE REVIEWS | DRUG DISCOVERY VOLUME 10 | JULY 2011 | 507
ANALYSIS
nrd_3480_jul11.indd 507 22/06/2011 10:16
81. Morphy, R., Kay, C. & Rankovic, Z. From magic bullets to designed multiple ligands.
82. Morphy, R. & Rankovic, Z. The physicochemical challenges of designing multiple
ligands. J. Med. Chem. 49, 49614970 (2006).
83. Hopkins, A.L. et al. Design of non-nucleoside inhibitors of HIV-1 reverse transcriptase
with improved drug resistance properties. 1. J. Med. Chem. 47, 59125922 (2004).
84. Hann, M.M., Leach, A.R. & Harper, G. Molecular complexity and its impact on the
probability of nding leads for drug discovery. J. Chem. Inf. Comput. Sci. 41,
856864 (2001).
85. Leach, A.R., Hann, M.M., Burrows, J.N. & Griffen, E.J. Fragment screening: an
introduction. Mol. Biosyst. 2, 430446 (2006).
86. Hopkins, A.L., Groom, C.R. & Alex, A. Ligand efciency: a useful metric for lead
selection. Drug Discov. Today 9, 430431 (2004).
87. Morphy, R. & Rankovic, Z. Fragments, network biology and designing multiple
ligands. Drug Discov. Today 12, 156160 (2007).
88. Neumann, T., Junker, H.D., Schmidt, K. & Sekul, R. SPR-based fragment screening:
advantages and applications. Curr. Top. Med. Chem. 7, 16301642 (2007).
89. Ha ma la inen, M.D. et al. Label-free primary screening and afnity ranking of fragment
libraries using parallel analysis of protein panels. J. Biomol. Screen. 13, 202209
(2008).
90. Paolini, G.V., Shapland, R.H., van Hoorn, W.P., Mason, J.S. & Hopkins, A.L. Global
mapping of pharmacological space. Nat. Biotechnol. 24, 805815 (2006).
91. Wermuth, C.G. Selective optimization of side activities: another way for drug
discovery. J. Med. Chem. 47, 13031314 (2004).
92. Wermuth, C.G. Selective optimization of side activities: the SOSA approach. Drug
Discov. Today 11, 160164 (2006).
93. Millan, M. Multi-target strategies for the improved treatment of depressive states:
conceptual foundations and neuronal substrates, drug discovery and therapeutic
application. Pharmacol. Ther. 110, 135370 (2006).
94. Ji, H.F. et al. Distribution patterns of small-molecule ligands in the protein universe
and implications for origin of life and drug discovery. Genome Biol. 8, R176 (2007).
95. Park, K. & Kim, D. Binding similarity network of ligand. Proteins 71, 960971
(2008).
96. Kuhn, M., von Mering, C., Campillos, M., Jensen, L.J. & Bork, P. STITCH: interaction
networks of chemicals and proteins. Nucleic Acids Res. 36, D684D688 (2008).
97. Gu nther, S. et al. SuperTarget and Matador: resources for exploring drug-target
relationships. Nucleic Acids Res. 36, D919D922 (2008).
98. Spiro, Z., Kovacs, I.A. & Csermely, P. Drug-therapy networks and the prediction of
novel drug targets. J. Biol. 7, 20 (2008).
99. Bonchev, D. & Buck, G.A. From molecular to biological structure and back. J. Chem.
Inf. Model. 47, 909917 (2007).
100. Keiser, M.J. et al. Relating protein pharmacology by ligand chemistry. Nat. Biotech-
nol. 25, 197206 (2007).
101. Hert, J., Keiser, M., Irwin, J., Oprea, T. & Shoichet, B. Quantifying the relationships
among drug classes. J. Chem. Inf. Model. 48, 755765 (2008).
102. Yamanishi, Y., Araki, M., Gutteridge, A., Honda, W. & Kanehisa, M. Prediction of
drug-target interaction networks from the integration of chemical and genomic
spaces. Bioinformatics 24, i232i240 (2008).
103. Kuhn, M., Campillos, M., Gonza lez, P., Jensen, L.J. & Bork, P. Large-scale prediction
of drugtarget relationships. FEBS Lett. 582, 12831289 (2008).
104. Campillos, M., Kuhn, M., Gavin, A.C., Jensen, L.J. & Bork, P. Drug target identica-
tion using side-effect similarity. Science 321, 263266 (2008).
105. Scott, M.S. & Barton, G.J. Probabilistic prediction and ranking of human protein-
protein interactions. BMC Bioinformatics 8, 239 (2007).
106. Fliri, A.F., Loging, W.T., Thadeio, P.F. & Volkmann, R.A. Biospectra analysis: model
proteome characterizations for linking molecular structure and biological response.
J. Med. Chem. 48, 69186925 (2005).
107. Lamb, J. et al. The connectivity map: using gene-expression signatures to
connect small molecules, genes, and disease. Science 313, 19291935
(2006).
108. Davies, J.R., Jackson, R.M., Mardia, K.V. & Taylor, C.C. The Poisson index: a new
probabilistic model for protein ligand binding site similarity. Bioinformatics 23,
30013008 (2007).
109. Baroni, M., Cruciani, G., Sciabola, S., Perruccio, F. & Mason, J.S. A common
reference framework for analyzing/comparing proteins and ligands. Fingerprints for
ligands and proteins (FLAP): theory and application. J. Chem. Inf. Model. 47,
279294 (2007).
110. Zhang, Z. & Grigorov, M.G. Similarity networks of protein binding sites. Proteins 62,
470478 (2006).
111. Zhang, X., Crespo, A. & Ferna ndez, A. Turning promiscuous kinase inhibitors into
safer drugs. Trends Biotechnol. 26, 295301 (2008).
112. Aronov, A.M., McClain, B., Moody, C.S. & Murcko, M.A. Kinase-likeness and kinase-
privileged fragments: toward virtual polypharmacology. J. Med. Chem. 51,
12141222 (2008).
113. Jenwitheesuk, E., Horst, J.A., Rivas, K.L., Van Voorhis, W.C. & Samudrala, R. Novel
paradigms for drug discovery: computational multitarget screening. Trends Pharma-
col. Sci. 29, 6271 (2008).
114. Van Gestel, S. & Schuermans, V. Thirty-three years of drug discovery and research
with Dr. Paul Janssen. Drug Dev. Res. 8, 113 (1986).
115. Overington, J.P., Al-Lazikani, B. & Hopkins, A.L. How many drug targets are there?
Nat. Rev. Drug Discov. 5, 993996 (2006).
116. Fabian, M.A. et al. A small moleculekinase interaction map for clinical kinase
inhibitors. Nat. Biotechnol. 23, 329336 (2005).
117. Shannon, P. et al. Cytoscape: a software environment for integrated
models of biomolecular interaction networks. Genome Res. 13, 24982504
(2003).
RE VI E W
Investment in drug research and development (R&D)
has increased substantially in recent decades, but the
annual number of truly innovative new medicines
approved by the US Food and Drug Administration
(FDA) has not increased accordingly, and attrition rates
are very high
1
. Indeed, in a recent analysis
2
it was noted
that without a dramatic improvement in R&D produc-
tivity, the pharmaceutical industry cannot sustain suf-
ficient innovation to replace the loss of revenues due to
patent expirations for successful products.
The authors of this analysis
2
also considered R&D pro-
ductivity in two dimensions: efficiency and effectiveness.
R&D efficiency represents the ability to translate inputs
(such as ideas, investment and effort) into defined out-
puts (such as milestones that represent resolved uncer-
tainties), whereas R&D effectiveness can be considered as
the ability to produce outputs with certain intended and
desired qualities. A key efficiency variable for increased
productivity is the probability of technical success. If the
probability of technical success could be increased (by
reducing attrition) for any given drug candidate or, ide-
ally, for a portfolio of drug candidates, then productivity
would increase accordingly. The authors also suggested
that target selection may be one of the most important
determinants of attrition and overall R&D productivity
2
.
Since the dawn of the genomics era in the 1990s, the
main focus of drug discovery has been on drug targets,
which are typically proteins that appear to have a key
role in disease pathogenesis
35
. Modification of target
activity provides a rational basis for the discovery of
new medicines; a target-centric approach provides a
specific biological hypothesis to be tested and a starting
point for the identification of molecules to do this with.
Tremendous advances have been made in the develop-
ment of new tools to identify targets (for example, RNA
interference) and compounds that interact with these
targets (for example, high-throughput target-based
screening assays that are applicable to key protein fami-
lies such as G protein-coupled receptors and kinases).
Structure-based tools that can be used to aid lead
identification and optimization for some targets have
also been developed, including X-ray crystallography
and computational modelling and screening (virtual
screening).
However, despite the power of these tools to identify
potential drug candidates, R&D productivity remains a
crucial challenge for the pharmaceutical industry, which
raises questions about the possible limitations of a tar-
get-centric approach to drug discovery. Indeed, before
the introduction of target-based approaches, drug dis-
covery was driven primarily by phenotypic assays, often
with limited knowledge of the molecular mechanisms
of disease. Nevertheless, the pharmaceutical industry
was successful in the discovery and development of new
*Roche Palo Alto,
3431 Hillview Avenue,
Palo Alto, California 94304,
USA.
iRND3 (Institute for Rare and

Neglected Diseases Drug
Discovery), 951 Old County
Road, PMB 316, Belmont,
California 940022760, USA.
Correspondence to D.C.S.
email david.swinney@irnd3.
org
doi:10.1038/nrd3480
How were new medicines discovered?
David C.Swinney*
and Jason Anthony*

Abstract | Preclinical strategies that are used to identify potential drug candidates include
target-based screening, phenotypic screening, modification of natural substances and
biologic-based approaches. To investigate whether some strategies have been more
successful than others in the discovery of new drugs, we analysed the discovery strategies
and the molecular mechanism of action (MMOA) for new molecular entities and new
biologics that were approved by the US Food and Drug Administration between 1999 and
2008. Out of the 259 agents that were approved, 75 were first-in-class drugs with new
MMOAs, and out of these, 50 (67%) were small molecules and 25 (33%) were biologics.
The results also show that the contribution of phenotypic screening to the discovery of
first-in-class small-molecule drugs exceeded that of target-based approaches with 28
and 17 of these drugs coming from the two approaches, respectively in an era in which
the major focus was on target-based approaches. We postulate that a target-centric
approach for first-in-class drugs, without consideration of an optimal MMOA, may
contribute to the current high attrition rates and low productivity in pharmaceutical
research and development.
ANALYSIS
nrd_3480_jul11.indd 507 22/06/2011 10:16
81. Morphy, R., Kay, C. & Rankovic, Z. From magic bullets to designed multiple ligands.
82. Morphy, R. & Rankovic, Z. The physicochemical challenges of designing multiple
ligands. J. Med. Chem. 49, 49614970 (2006).
83. Hopkins, A.L. et al. Design of non-nucleoside inhibitors of HIV-1 reverse transcriptase
with improved drug resistance properties. 1. J. Med. Chem. 47, 59125922 (2004).
84. Hann, M.M., Leach, A.R. & Harper, G. Molecular complexity and its impact on the
probability of nding leads for drug discovery. J. Chem. Inf. Comput. Sci. 41,
856864 (2001).
85. Leach, A.R., Hann, M.M., Burrows, J.N. & Griffen, E.J. Fragment screening: an
introduction. Mol. Biosyst. 2, 430446 (2006).
86. Hopkins, A.L., Groom, C.R. & Alex, A. Ligand efciency: a useful metric for lead
selection. Drug Discov. Today 9, 430431 (2004).
87. Morphy, R. & Rankovic, Z. Fragments, network biology and designing multiple
ligands. Drug Discov. Today 12, 156160 (2007).
88. Neumann, T., Junker, H.D., Schmidt, K. & Sekul, R. SPR-based fragment screening:
advantages and applications. Curr. Top. Med. Chem. 7, 16301642 (2007).
89. Ha ma la inen, M.D. et al. Label-free primary screening and afnity ranking of fragment
libraries using parallel analysis of protein panels. J. Biomol. Screen. 13, 202209
(2008).
90. Paolini, G.V., Shapland, R.H., van Hoorn, W.P., Mason, J.S. & Hopkins, A.L. Global
mapping of pharmacological space. Nat. Biotechnol. 24, 805815 (2006).
91. Wermuth, C.G. Selective optimization of side activities: another way for drug
discovery. J. Med. Chem. 47, 13031314 (2004).
92. Wermuth, C.G. Selective optimization of side activities: the SOSA approach. Drug
Discov. Today 11, 160164 (2006).
93. Millan, M. Multi-target strategies for the improved treatment of depressive states:
conceptual foundations and neuronal substrates, drug discovery and therapeutic
application. Pharmacol. Ther. 110, 135370 (2006).
94. Ji, H.F. et al. Distribution patterns of small-molecule ligands in the protein universe
and implications for origin of life and drug discovery. Genome Biol. 8, R176 (2007).
95. Park, K. & Kim, D. Binding similarity network of ligand. Proteins 71, 960971
(2008).
96. Kuhn, M., von Mering, C., Campillos, M., Jensen, L.J. & Bork, P. STITCH: interaction
networks of chemicals and proteins. Nucleic Acids Res. 36, D684D688 (2008).
97. Gu nther, S. et al. SuperTarget and Matador: resources for exploring drug-target
relationships. Nucleic Acids Res. 36, D919D922 (2008).
98. Spiro, Z., Kovacs, I.A. & Csermely, P. Drug-therapy networks and the prediction of
novel drug targets. J. Biol. 7, 20 (2008).
99. Bonchev, D. & Buck, G.A. From molecular to biological structure and back. J. Chem.
Inf. Model. 47, 909917 (2007).
100. Keiser, M.J. et al. Relating protein pharmacology by ligand chemistry. Nat. Biotech-
nol. 25, 197206 (2007).
101. Hert, J., Keiser, M., Irwin, J., Oprea, T. & Shoichet, B. Quantifying the relationships
among drug classes. J. Chem. Inf. Model. 48, 755765 (2008).
102. Yamanishi, Y., Araki, M., Gutteridge, A., Honda, W. & Kanehisa, M. Prediction of
drug-target interaction networks from the integration of chemical and genomic
spaces. Bioinformatics 24, i232i240 (2008).
103. Kuhn, M., Campillos, M., Gonza lez, P., Jensen, L.J. & Bork, P. Large-scale prediction
of drugtarget relationships. FEBS Lett. 582, 12831289 (2008).
104. Campillos, M., Kuhn, M., Gavin, A.C., Jensen, L.J. & Bork, P. Drug target identica-
tion using side-effect similarity. Science 321, 263266 (2008).
105. Scott, M.S. & Barton, G.J. Probabilistic prediction and ranking of human protein-
protein interactions. BMC Bioinformatics 8, 239 (2007).
106. Fliri, A.F., Loging, W.T., Thadeio, P.F. & Volkmann, R.A. Biospectra analysis: model
proteome characterizations for linking molecular structure and biological response.
J. Med. Chem. 48, 69186925 (2005).
107. Lamb, J. et al. The connectivity map: using gene-expression signatures to
connect small molecules, genes, and disease. Science 313, 19291935
(2006).
108. Davies, J.R., Jackson, R.M., Mardia, K.V. & Taylor, C.C. The Poisson index: a new
probabilistic model for protein ligand binding site similarity. Bioinformatics 23,
30013008 (2007).
109. Baroni, M., Cruciani, G., Sciabola, S., Perruccio, F. & Mason, J.S. A common
reference framework for analyzing/comparing proteins and ligands. Fingerprints for
ligands and proteins (FLAP): theory and application. J. Chem. Inf. Model. 47,
279294 (2007).
110. Zhang, Z. & Grigorov, M.G. Similarity networks of protein binding sites. Proteins 62,
470478 (2006).
111. Zhang, X., Crespo, A. & Ferna ndez, A. Turning promiscuous kinase inhibitors into
safer drugs. Trends Biotechnol. 26, 295301 (2008).
112. Aronov, A.M., McClain, B., Moody, C.S. & Murcko, M.A. Kinase-likeness and kinase-
privileged fragments: toward virtual polypharmacology. J. Med. Chem. 51,
12141222 (2008).
113. Jenwitheesuk, E., Horst, J.A., Rivas, K.L., Van Voorhis, W.C. & Samudrala, R. Novel
paradigms for drug discovery: computational multitarget screening. Trends Pharma-
col. Sci. 29, 6271 (2008).
114. Van Gestel, S. & Schuermans, V. Thirty-three years of drug discovery and research
with Dr. Paul Janssen. Drug Dev. Res. 8, 113 (1986).
115. Overington, J.P., Al-Lazikani, B. & Hopkins, A.L. How many drug targets are there?
Nat. Rev. Drug Discov. 5, 993996 (2006).
116. Fabian, M.A. et al. A small moleculekinase interaction map for clinical kinase
inhibitors. Nat. Biotechnol. 23, 329336 (2005).
117. Shannon, P. et al. Cytoscape: a software environment for integrated
models of biomolecular interaction networks. Genome Res. 13, 24982504
(2003).
RE VI E W
First published in Nature Reviews Drug Discovery 10, 507519 (July 2011) | doi:10.1038/nrd3480
Table 1 | First-in-class small-molecule new molecular entities approved by the FDA: 19992008
Drug (trade name; company) Therapeutic area Target type Molecular mechanism of action Refs
Discovered through phenotypic screening
Aripiprazole (Abilify; Bristol-Myers
Squibb/Otsuka Pharmaceutical)
CNS Receptor Conformational/partial agonist 74,75,
8084
Azacitidine (Vidaza; Celgene/Pfizer) Cancer Enyzme Irreversible inhibition 69,85
Caspofungin (Cancidas; Merck) Infectious disease Enzyme Noncompetitive inhibition 71,86
Cilostazol (Pletal; Otsuka) Cardiovascular Enzyme Inhibition 87
Cinacalcet (Sensipar; Amgen) Metabolic Receptor Allosteric activator 29
Daptomycin (Cubicin; Cubist) Infectious disease NA (disrupts bacterial
membrane)
Unknown 88
Docosanol (Abreva; Avanir
Pharmaceuticals)
Infectious disease Unknown Unknown 8992
Ezetimibe (Zetia; Merck) Cardiovascular Transporter Slow binding kinetics 30
Fulvestrant (Faslodex; AstraZeneca) Cancer Receptor Antagonist-induced degradation 47,93,94
Levetiracetam (Keppra; UCB Pharma) CNS Unknown Unknown 95
Linezolid (Zyvox; Pfizer) Infectious disease Enzyme Conformational 28,96,97
Lubiprostone (Amitiza; Sucampo
Pharmaceuticals)
Gastrointestinal Unknown Unknown 98100
Memantine (Namenda; Forest) CNS Receptor Uncompetitive and fast
binding kinetics
101103
Miglustat (Zavesca; Actelion) Rare diseases Enzyme Reversible inhibition 104,105
Nateglinide
(Fastic; Novartis/Astellas)
Metabolic Unknown Fast binding kinetics 106108
Nelarabine (Arranon;
GlaxoSmithKline)
Cancer DNA (nucleoside
analogue)
Nucleotide chain termination 109113
Nitazoxanide (Alinia; Roche) Infectious disease Enzyme Irreversible and redox 78,79
Nitisinone (Orfadin; Syngenta) Rare diseases Enzyme Irreversible 114116
Pemirolast (Alamast; Senten) Immune modulation Unknown Unknown 117
Ranolazine (Ranexa; Gilead) Cardiovascular Unknown Unknown 118121
Retapamulin (Altabax;
GlaxoSmithKline)
Infectious disease Enzyme Allosteric inhibitor 122
Rufinamide (Inovelon; Novartis) CNS Unknown Unknown 123,124
Sinecatechins
(Veregen; Medigene)
Sirolimus (Rapamune; Pfizer) Immune modulation Enzyme Conformational/inhibition 70,126
Varenicline (Chantix; Pfizer) CNS Ion channel Conformational/partial agonist 76
Vorinostat (Zolinza; Merck) Cancer Enzyme Equilibrium kinetics 127,128
Ziconotide (Prialt; Elan
Pharmaceuticals)
Pain and/or CNS Ion channel Equilibrium kinetics 31
Zonisamide (Excegran; Dainippon
Pharmaceuticals)
CNS Unknown Unknown 129
Discovered through target-based screening
Aliskiren (Tekturna; Novartis) Cardiovascular Enzyme Equilibrium binding 38,130
Aprepitant (Emend; Merck) Gastrointestinal Receptor Slow binding kinetics 46
Bortezomib (Velcade; Millenium
Pharmaceuticals)
Cancer Enzyme Equilibrium binding 131,132
Bosentan (Tracleer; Actelion) Cardiovascular Receptor Equilibrium binding 37
Conivaptan
(Vaprisol; Astellas Pharma)
Metabolic Receptor Equilibrium binding 133
Eltrombopag (Promacta;
GlaxoSmithKline)
Immune Receptor Noncompetitive agonist 36
Gefitinib (Iressa; AstraZeneca) Cancer Enzyme Stabilize inactive conformation 41,42
ANALYSI S
nrd_3480_jul11.indd 509 22/06/2011 10:16
New molecular entities
(NME). A medication
form in the United States.
innovative medicines; it has therefore been suggested
that the more limited use of phenotypic screening in
recent years has contributed to the current lack of suc-
cess in drug R&D
6,7
.
These two different approaches to drug discovery
target-based screening and phenotypic screening
each have advantages and disadvantages. The strengths
of the target-based approach include the ability to apply
molecular and chemical knowledge to investigate spe-
cific molecular hypotheses, and the ability to apply both
small-molecule screening strategies (which can often be
achieved using high-throughput formats) and biologic-
based approaches, such as identifying monoclonal anti-
bodies. A disadvantage of the target-based approach is
that the solution to the specific molecular hypotheses
may not be relevant to the disease pathogenesis or pro-
vide a sufficient therapeutic index.
A strength of the phenotypic approach is that the
assays do not require prior understanding of the molecu-
lar mechanism of action (MMOA), and activity in such
assays might be translated into therapeutic impact in a
given disease state more effectively than in target-based
assays, which are often more artificial. A disadvantage
of phenotypic screening approaches is the challenge
of optimizing the molecular properties of candidate
drugs without the design parameters provided by prior
knowledge of the MMOA. An additional challenge is
to effectively incorporate new screening technologies
into phenotypic screening approaches, which is impor-
tant for addressing the traditional limitation of some
of these assays: a considerably lower throughput than
target-based assays.
In order to gain a better understanding of the factors
that could contribute to the high attrition rates, and to
provide insights that might help to reduce attrition and
increase R&D productivity, we decided to investigate the
approaches that were used in the discovery of recently
introduced medicines. To achieve this, we analysed the
characteristics of the new molecular entities (NMEs) and
new therapeutic biologics that were approved by the
FDA during the 10-year period between 1999 and 2008
by examining the discovery approach, the MMOA and
whether the drug was first in itsclass.
Data and analysis
Numbers of NMEs. In the 10-year period between 1999
and 2008, the FDA approved 183 small-molecule drugs,
20 imaging agents and 56 new therapeutic biologics (259
agents overall). Out of these, 75 drugs were identified as
first-in-class or with novel MMOAs based on the infor-
mation provided in the product labels on the FDA web-
site (see the Drugs@FDA website), and primary research
and review publications (TABLE 1; Supplementary infor-
mation S1 (table)). The specific sources for each drug are
referenced in TABLE 1 and in Supplementary information
S2 (box).
Discovery approaches. We divided the list of 259
agents into three general categories: first-in-class drugs
(75 drugs), follower drugs (164 drugs) and imaging
agents (20 agents; these were not further analysed). A
list of all the drugs and their classification is provided
in Supplementary informationS1 (table) and a brief
description of the discovery history of first-in-class drugs
is provided in Supplementary information S2 (box). We
categorized the method of discovery of each new drug as
target-based, phenotypic-based, modification of a natural
substance, biologic-based or other (see Supplementary
informationS1 (table)). Overall, 100 NMEs were dis-
covered using target-based approaches, 58 NMEs were
discovered using phenotypic-based approaches, 18
NMEs were based on modifications of natural sub-
stances and 56 of the agents were biologics. All of the
biologics can be considered to have been discovered
using a target-based approach, and the main focus of
our analysis is on the methods of discovery for small-
molecule first-in-class NMEs and follower NMEs; that
is, small molecules that are in the same class as a previ-
ously approvedNME.
MMOA.The MMOA of the NMEs was analysed because
the limitations of a target-based approach with respect
to the MMOA have been highlighted
810
, and because
the MMOA is a characteristic of drugs that has received
less attention with regard to its connection to attrition.
For the purpose of this article, MMOA is defined as the
biochemical mechanism through which the structural
interactions between the drug and its target(s) result in
a functional response
1012
, which is important in both
drug efficacy and safety (BOX1). The MMOA can affect
how efficiently a binding interaction is coupled to the
functional response, which can be assessed by consider-
ing biochemical efficiency (BOX2).
For instance, resistance to the ATP-competitive
kinase inhibitors gefitinib and erlotinib which target
the epidermal growth factor receptor (EGFR) kinase
has been shown to be due to mutations that alter
the ATP binding site in such a way that they increase
the affinity of the EGFR kinase domain for ATP. The
functional consequence of these resistance mutations
is therefore to enable ATP to compete more effec-
tively with gefitinib and erlotinib
12,13
. This provides an
explanation for the mechanism of resistance to these
rapidly reversible ATP-competitive inhibitors, and also
provides an explanation as to why irreversible covalent
binding inhibitors overcome this resistance
13
.
An example of how the therapeutic utility of drugs
that function through interaction with a receptor is
influenced by their MMOA is provided by the tissue-
selective functional effects of the selective oestrogen
receptor modulators (SERMs), which are mediated
by SERM-induced structural changes in the oestrogen
receptor
14
. Binding to the receptor initiates a series of
molecular events, which culminate in the activation
or repression of specific genes. The SERMs tamoxifen
and raloxifene bind at the same site within the core of
the ligand-binding domain, but with different bind-
ing modes that are translated into distinct conforma-
tions of the transactivation domain of the receptor.
Transcriptional regulation of the oestrogen receptor
is a complex process that involves the participa-
tion of co-activators and co-repressors, and the
ANALYSI S
508 | JULY 2011 | VOLUME 10 www.nature.com/reviews/drugdisc
nrd_3480_jul11.indd 508 22/06/2011 10:16
Table 1 | First-in-class small-molecule new molecular entities approved by the FDA: 19992008
Discovered through phenotypic screening
Aripiprazole (Abilify; Bristol-Myers
Squibb/Otsuka Pharmaceutical)
CNS Receptor Conformational/partial agonist 74,75,
8084
Azacitidine (Vidaza; Celgene/Pfizer) Cancer Enyzme Irreversible inhibition 69,85
Caspofungin (Cancidas; Merck) Infectious disease Enzyme Noncompetitive inhibition 71,86
Cilostazol (Pletal; Otsuka) Cardiovascular Enzyme Inhibition 87
Cinacalcet (Sensipar; Amgen) Metabolic Receptor Allosteric activator 29
Daptomycin (Cubicin; Cubist) Infectious disease NA (disrupts bacterial
membrane)
Unknown 88
Docosanol (Abreva; Avanir
Pharmaceuticals)
Ezetimibe (Zetia; Merck) Cardiovascular Transporter Slow binding kinetics 30
Fulvestrant (Faslodex; AstraZeneca) Cancer Receptor Antagonist-induced degradation 47,93,94
Levetiracetam (Keppra; UCB Pharma) CNS Unknown Unknown 95
Linezolid (Zyvox; Pfizer) Infectious disease Enzyme Conformational 28,96,97
Lubiprostone (Amitiza; Sucampo
Pharmaceuticals)
Gastrointestinal Unknown Unknown 98100
Memantine (Namenda; Forest) CNS Receptor Uncompetitive and fast
binding kinetics
101103
Miglustat (Zavesca; Actelion) Rare diseases Enzyme Reversible inhibition 104,105
Nateglinide
(Fastic; Novartis/Astellas)
Metabolic Unknown Fast binding kinetics 106108
Nelarabine (Arranon;
GlaxoSmithKline)
Cancer DNA (nucleoside
analogue)
Nucleotide chain termination 109113
Nitazoxanide (Alinia; Roche) Infectious disease Enzyme Irreversible and redox 78,79
Nitisinone (Orfadin; Syngenta) Rare diseases Enzyme Irreversible 114116
Pemirolast (Alamast; Senten) Immune modulation Unknown Unknown 117
Ranolazine (Ranexa; Gilead) Cardiovascular Unknown Unknown 118121
Retapamulin (Altabax;
GlaxoSmithKline)
Infectious disease Enzyme Allosteric inhibitor 122
Rufinamide (Inovelon; Novartis) CNS Unknown Unknown 123,124
Sinecatechins
(Veregen; Medigene)
Sirolimus (Rapamune; Pfizer) Immune modulation Enzyme Conformational/inhibition 70,126
Varenicline (Chantix; Pfizer) CNS Ion channel Conformational/partial agonist 76
Vorinostat (Zolinza; Merck) Cancer Enzyme Equilibrium kinetics 127,128
Ziconotide (Prialt; Elan
Pharmaceuticals)
Pain and/or CNS Ion channel Equilibrium kinetics 31
Zonisamide (Excegran; Dainippon
Pharmaceuticals)
CNS Unknown Unknown 129
Discovered through target-based screening
Aliskiren (Tekturna; Novartis) Cardiovascular Enzyme Equilibrium binding 38,130
Aprepitant (Emend; Merck) Gastrointestinal Receptor Slow binding kinetics 46
Bortezomib (Velcade; Millenium
Pharmaceuticals)
Cancer Enzyme Equilibrium binding 131,132
Bosentan (Tracleer; Actelion) Cardiovascular Receptor Equilibrium binding 37
Conivaptan
(Vaprisol; Astellas Pharma)
Metabolic Receptor Equilibrium binding 133
Eltrombopag (Promacta;
GlaxoSmithKline)
Immune Receptor Noncompetitive agonist 36
Gefitinib (Iressa; AstraZeneca) Cancer Enzyme Stabilize inactive conformation 41,42
ANALYSI S
nrd_3480_jul11.indd 509 22/06/2011 10:16
New molecular entities
(NME). A medication
form in the United States.
innovative medicines; it has therefore been suggested
that the more limited use of phenotypic screening in
recent years has contributed to the current lack of suc-
cess in drug R&D
6,7
.
These two different approaches to drug discovery
target-based screening and phenotypic screening
each have advantages and disadvantages. The strengths
of the target-based approach include the ability to apply
molecular and chemical knowledge to investigate spe-
cific molecular hypotheses, and the ability to apply both
small-molecule screening strategies (which can often be
achieved using high-throughput formats) and biologic-
based approaches, such as identifying monoclonal anti-
bodies. A disadvantage of the target-based approach is
that the solution to the specific molecular hypotheses
may not be relevant to the disease pathogenesis or pro-
vide a sufficient therapeutic index.
A strength of the phenotypic approach is that the
assays do not require prior understanding of the molecu-
lar mechanism of action (MMOA), and activity in such
assays might be translated into therapeutic impact in a
given disease state more effectively than in target-based
assays, which are often more artificial. A disadvantage
of phenotypic screening approaches is the challenge
of optimizing the molecular properties of candidate
drugs without the design parameters provided by prior
knowledge of the MMOA. An additional challenge is
to effectively incorporate new screening technologies
into phenotypic screening approaches, which is impor-
tant for addressing the traditional limitation of some
of these assays: a considerably lower throughput than
target-based assays.
In order to gain a better understanding of the factors
that could contribute to the high attrition rates, and to
provide insights that might help to reduce attrition and
increase R&D productivity, we decided to investigate the
approaches that were used in the discovery of recently
introduced medicines. To achieve this, we analysed the
characteristics of the new molecular entities (NMEs) and
new therapeutic biologics that were approved by the
FDA during the 10-year period between 1999 and 2008
by examining the discovery approach, the MMOA and
whether the drug was first in itsclass.
Data and analysis
Numbers of NMEs. In the 10-year period between 1999
and 2008, the FDA approved 183 small-molecule drugs,
20 imaging agents and 56 new therapeutic biologics (259
agents overall). Out of these, 75 drugs were identified as
first-in-class or with novel MMOAs based on the infor-
mation provided in the product labels on the FDA web-
site (see the Drugs@FDA website), and primary research
and review publications (TABLE 1; Supplementary infor-
mation S1 (table)). The specific sources for each drug are
referenced in TABLE 1 and in Supplementary information
S2 (box).
Discovery approaches. We divided the list of 259
agents into three general categories: first-in-class drugs
(75 drugs), follower drugs (164 drugs) and imaging
agents (20 agents; these were not further analysed). A
list of all the drugs and their classification is provided
in Supplementary informationS1 (table) and a brief
description of the discovery history of first-in-class drugs
is provided in Supplementary information S2 (box). We
categorized the method of discovery of each new drug as
target-based, phenotypic-based, modification of a natural
substance, biologic-based or other (see Supplementary
informationS1 (table)). Overall, 100 NMEs were dis-
covered using target-based approaches, 58 NMEs were
discovered using phenotypic-based approaches, 18
NMEs were based on modifications of natural sub-
stances and 56 of the agents were biologics. All of the
biologics can be considered to have been discovered
using a target-based approach, and the main focus of
our analysis is on the methods of discovery for small-
molecule first-in-class NMEs and follower NMEs; that
is, small molecules that are in the same class as a previ-
ously approvedNME.
MMOA.The MMOA of the NMEs was analysed because
the limitations of a target-based approach with respect
to the MMOA have been highlighted
810
, and because
the MMOA is a characteristic of drugs that has received
less attention with regard to its connection to attrition.
For the purpose of this article, MMOA is defined as the
biochemical mechanism through which the structural
interactions between the drug and its target(s) result in
a functional response
1012
, which is important in both
drug efficacy and safety (BOX1). The MMOA can affect
how efficiently a binding interaction is coupled to the
functional response, which can be assessed by consider-
ing biochemical efficiency (BOX2).
For instance, resistance to the ATP-competitive
kinase inhibitors gefitinib and erlotinib which target
the epidermal growth factor receptor (EGFR) kinase
has been shown to be due to mutations that alter
the ATP binding site in such a way that they increase
the affinity of the EGFR kinase domain for ATP. The
functional consequence of these resistance mutations
is therefore to enable ATP to compete more effec-
tively with gefitinib and erlotinib
12,13
. This provides an
explanation for the mechanism of resistance to these
rapidly reversible ATP-competitive inhibitors, and also
provides an explanation as to why irreversible covalent
binding inhibitors overcome this resistance
13
.
An example of how the therapeutic utility of drugs
that function through interaction with a receptor is
influenced by their MMOA is provided by the tissue-
selective functional effects of the selective oestrogen
receptor modulators (SERMs), which are mediated
by SERM-induced structural changes in the oestrogen
receptor
14
. Binding to the receptor initiates a series of
molecular events, which culminate in the activation
or repression of specific genes. The SERMs tamoxifen
and raloxifene bind at the same site within the core of
the ligand-binding domain, but with different bind-
ing modes that are translated into distinct conforma-
tions of the transactivation domain of the receptor.
Transcriptional regulation of the oestrogen receptor
is a complex process that involves the participa-
tion of co-activators and co-repressors, and the
ANALYSI S
nrd_3480_jul11.indd 508 22/06/2011 10:16
worth noting that several of these NMEs (for example,
nelarabine, azacitidine and nitazoxanide) were initially
described decades before their approval and before the
development of new molecular screening approaches.
Many of these NMEs were also derived from natural
substances, including the nucleoside analogues nelara-
bine and azacitidine, the PGE1 derivative lubiprostone
and the fatty acid docosanol. Ziconotide, sirolimus and
retapamulin were derived from natural products.
NMEs that were developed as synthetic and/or modi-
fied versions of natural substances, or discovered by
screening such substances. A small fraction of the
first-in-class NMEs (5 out of 75) were developed as
synthetic versions of natural substances (that were
sometimes slightly modified), including the modified
heparin fondaparinux, the porphyrin verteporfin, the
biopterin cofactor sapropterin, the porphyrin precur-
sor aminolevulinic acid and the acetylated homotaurine

Box 1 | Molecular mechanism of action
The molecular mechanism of action (MMOA) is defined here as the interaction between a drug and its target (or targets) that
creates a specific response. These specific molecular interactions link structure to function in such a manner as to provide a
therapeutically effective and safe response. In this context, the MMOA is differentiated from mechanism of action (MOA), which
describes the mechanism in the context of the physiological response such as antihistamines, anti-inflammatory, and so on.
There are many facets of this interaction that ultimately result in the desired therapeutic outcome. For example, the site
of interaction (allosteric or orthosteric), molecular descriptors of the binding interaction (such as affinity and binding
kinetics), the functional impact (for example, receptor agonism, modulation or antagonism) and the specificity of the
functional outcome (for example, activation of specific signalling pathways) all contribute to the MMOA and affect the
ultimate pharmacological response.
Possible MMOAs at a target are listed below, together with selected examples of drugs that act through these MMOAs.
Kinetic mechanisms
For kinetic mechanisms, the pharmacological response to the drug is primarily driven by binding kinetics and residence
time at the target
12,1719
.
Equilibrium binding. The response to the drug is represented by the equilibrium dissociation constant (K
i
) to the target.
The binding has sufficiently fast on and off rates (k
on
and k
off
) to allow equilibrium to be reached and is thereby sensitive to
competition with physiological substrates and/or ligands (for example, bosentan, an endothelin receptor antagonist; and
aliskiren, a renin inhibitor)
37,38,68
.
Slow kinetics. Non-equilibrium and irreversible mechanisms involve slow association and/or dissociation rates (k
on
and k
off
)
that do not allow equilibrium to be reached and are less sensitive to competition with physiological substrates and/or ligands
(for example, orlistat binds irreversibly to the active site serine of pancreatic lipase, azacitidine irreversibly binds to DNA
methyltransferases and candesartan has a slow dissociation rate from the angiotensin II receptor)
1720,35,63,69
.
Conformational mechanisms
For conformational mechanisms, binding of the drug to the target involves a conformational change in the target that
couples drug binding to a response (for example, sirolimus binds to the peptidylprolyl isomerase FKBP12, which stabilizes
a conformation that subsequently inhibits the kinase activity of mammalian target of rapamycin; and fulvestrant induces a
conformation of the oestrogen receptor that is subsequently degraded)
811,47,70
.
Noncompetitive inhibition and/or antagonism. This is a form of MMOA in which the drug binds to a site on the target that
is distinct from the physiological substrate- and/or ligand-binding site that results in an inhibition of the response (for
example, caspofungin is a noncompetitive inhibitor of 1,3--d-glucan synthase owing to the observation that its IC
50

(half-maximal inhibitory concentration) is not influenced by substrate concentrations)
68,71
.
Uncompetitive inhibition and/or antagonism. An uncompetitive MMOA is contingent on prior activation of the target by
the physiological effector (the substrate or the ligand). This means that the same amount of drug blocks higher concentrations
of the physiological effector to a greater degree than lower concentrations. For example, memantine is an uncompetitive
antagonist that binds only to the activated form of the NMDA receptor. The potency of the inhibition of the NMDA receptor by
memantine increases at higher concentrations of glutamate (the physiological ligand)
22,23,68
.
Full agonism. Maximal efficacy is produced following drug binding to the receptor and subsequent receptor activation (for
example, ramelteon mimics the activity of melatonin for the melatonin receptor through binding at the orthosteric site
with efficient coupling to activate specific signalling pathways)
72,73
.
Partial agonism. This is a form of MMOA in which only partial efficacy is produced following drug binding to the orthosteric
site on the receptor (for example, aripiprazole is a partial agonist of the dopamine D2 receptor and varenciline is a partial
agonist of the nicotinic acetylcholine receptors )
73,7476
.
Allosteric modulator. This mechanism involves regulation of the biological activity of the target by binding of a drug at a
site other than the binding site for the endogenous substrate and/or ligand (allosteric site) (for example, cinacalcet is an
allosteric modulator of the calcium receptor by binding to the allosteric site)
29,73
.
Redox mechanisms
Redox is short for reductionoxidation reactions in which the pharmacological response to the drug is a consequence of
electron transfer between the drug and a physiological target. For example, generation of hydroxyl radicals by verteporfin
is thought to contribute to its ability to damage cells, and the antiprotozoal activity of nitazoxanide is believed to be due
in part to interference with the pyruvateferredoxin oxidoreductase enzyme-dependent electron transfer reaction, which
is essential to anaerobic energy metabolism
7779
.
ANALYSI S
nrd_3480_jul11.indd 511 22/06/2011 10:16
different conformations presumably change the affinity
of the receptor for the interacting co-activators and co-
repressors. The change in co-repressor affinity alters
the composition of the distinct cellular co-regulatory
complexes that modulate the functional transcriptional
activity
14
.

The MMOA can also differentiate similar drugs with
respect to their therapeutic indications. At the structural
level, aspirin is an irreversible inhibitor of cyclooxyge-
nases, whereas ibuprofen and naproxen are reversible
inhibitors. All three molecules bind to cyclooxygenase
enzymes at the same substrate binding site. However,
the irreversible MMOA of aspirin differentiates its
functional use as an antiplatelet drug from the revers-
ible inhibitors, because this MMOA translates into a
long-lasting action of aspirin in platelets, as platelets do
not have the capacity to resynthesize new enzymes
15,16
.
There are many different biochemical features of
an MMOA through which molecular interactions can
contribute to a specific functional response. These
include residence time
10,1719
, irreversible binding
20
,
transient binding
21,22
, and uncompetitive
22,23
and non-
competitive
10
inhibition mechanisms (BOX 1). It has
been proposed that drugs should be activated by the
pathological state that they are intended to inhibit
22,23
.
Allosteric inhibition and activation are important for
the pharmacological modulation of many receptors and
channels
24,25
. Voltage- or frequency-dependent channel
blockade can also influence a selective pharmacological
response
26,27
. Given the importance of the MMOA to the
therapeutic effects of NMEs, we consider it further in
the following sections.
Discovery of first-in-class medicines
NMEs that were discovered through phenotypic screen-
ing. The 28 first-in-class small-molecule NMEs that
were discovered in phenotypic screens either came from
intentional targeting of a specific phenotype (25NMEs)
or through serendipity (3NMEs) (FIG.1). The inten-
tional approaches were based on assays that measured
a specific physiological phenomenon, with little under-
standing of the MMOA. In many cases, the newly dis-
covered molecules were subsequently used to identify
MMOAs for the physiological phenomena. For example,
the oxazolidinone antibiotics (such as linezolid) were
initially discovered as inhibitors of Gram-positive bac-
teria but were subsequently shown to be protein synthe-
sis inhibitors that target an early step in the binding of
N-formylmethionyl-tRNA to the ribosome
28
. This is also
illustrated by the calcium receptor allosteric activator cin-
acalcet
29
, the sterol transporter inhibitor ezetimibe
30
and
the N-type calcium channel blocker ziconotide
31
; these
drugs were initially discovered using phenotypic assays.
The majority of discoveries focused on using specific
chemical classes in which prior knowledge contributed
to matching the chemical class with the phenotype
for example, screening nucleoside analogues as poten-
tial anticancer and antiviral agents. Random library
screening was also successful for ezetimibe, linezolid,
pemirolast, retapamulin, rufinamide and sirolimus. An
additional approach was to use phenotypic screening
to identify new MMOAs for established targets, which
led to the discovery of the partial agonists aripiprazole
and varenicline, and the full antagonist fulvestrant (see
Supplementary information S2 (box) for details). It is
Table 1 cont. | First-in-class small-molecule new molecular entities approved by the FDA: 19992008
Imatinib (Gleevec; Novartis) Cancer Enzyme Stabilizes inactive conformation 49
Maraviroc (Celsentri; Pfizer) Infectious disease Receptor Conformational and/or allosteric 134137
Mifepristone (Mifeprex; Aventis
Pharma)
Reproductive Receptor Conformational antagonist 138141
Orlistat (Xenical; Roche) Metabolic Enzyme Irreversible 35,142
Raltegravir (Isentress; Merck) Infectious disease Enzyme Traps conformational state 39,40,143,
144
Ramelteon (Rozerem; Takeda
Pharmaceuticals)
CNS Receptor Equilibrium binding 72,145
Sitagliptin (Januvia; Merck) Metabolic Enzyme Equilibrium binding 33,146
Sorafenib (Nexavar; Bayer) Cancer Enzyme Conformation state-specific inhibition 44
Sunitinib (Sutent; Pfizer) Cancer Enzyme Conformation state-specific inhibition 147150
Zanamivir
(Relenza; GlaxoSmithKline)
Infectious disease Enzyme Equilibrium binding 34,151
Discovered based on natural substrate or natural substance
Acamprosate (Campral; Merck) CNS Ion channel Conformational channel modulator 152
Aminolevulinic acid (Levulan; Berlex) Dermatology NA (photosensitizer) Redox 153,154
Fondaparinux (Arixtra; Sanofi) Cardiovascular Enzyme Irreversible 155157
Sapropterin (Kuvan; BioMarin) Rare diseases Enzyme Cofactor 158161
Verteporfin (Visudyne; QLT) Ocular NA (photoreaction) Redox 77,162
CNS, central nervous system; FDA, US Food and Drug Administration; NA, not applicable.
ANALYSI S
nrd_3480_jul11.indd 510 22/06/2011 10:16
worth noting that several of these NMEs (for example,
nelarabine, azacitidine and nitazoxanide) were initially
described decades before their approval and before the
development of new molecular screening approaches.
Many of these NMEs were also derived from natural
substances, including the nucleoside analogues nelara-
bine and azacitidine, the PGE1 derivative lubiprostone
and the fatty acid docosanol. Ziconotide, sirolimus and
retapamulin were derived from natural products.
NMEs that were developed as synthetic and/or modi-
fied versions of natural substances, or discovered by
screening such substances. A small fraction of the
first-in-class NMEs (5 out of 75) were developed as
synthetic versions of natural substances (that were
sometimes slightly modified), including the modified
heparin fondaparinux, the porphyrin verteporfin, the
biopterin cofactor sapropterin, the porphyrin precur-
sor aminolevulinic acid and the acetylated homotaurine

Box 1 | Molecular mechanism of action
The molecular mechanism of action (MMOA) is defined here as the interaction between a drug and its target (or targets) that
creates a specific response. These specific molecular interactions link structure to function in such a manner as to provide a
therapeutically effective and safe response. In this context, the MMOA is differentiated from mechanism of action (MOA), which
describes the mechanism in the context of the physiological response such as antihistamines, anti-inflammatory, and so on.
There are many facets of this interaction that ultimately result in the desired therapeutic outcome. For example, the site
of interaction (allosteric or orthosteric), molecular descriptors of the binding interaction (such as affinity and binding
kinetics), the functional impact (for example, receptor agonism, modulation or antagonism) and the specificity of the
functional outcome (for example, activation of specific signalling pathways) all contribute to the MMOA and affect the
ultimate pharmacological response.
Possible MMOAs at a target are listed below, together with selected examples of drugs that act through these MMOAs.
Kinetic mechanisms
For kinetic mechanisms, the pharmacological response to the drug is primarily driven by binding kinetics and residence
time at the target
12,1719
.
Equilibrium binding. The response to the drug is represented by the equilibrium dissociation constant (K
i
) to the target.
The binding has sufficiently fast on and off rates (k
on
and k
off
) to allow equilibrium to be reached and is thereby sensitive to
competition with physiological substrates and/or ligands (for example, bosentan, an endothelin receptor antagonist; and
aliskiren, a renin inhibitor)
37,38,68
.
Slow kinetics. Non-equilibrium and irreversible mechanisms involve slow association and/or dissociation rates (k
on
and k
off
)
that do not allow equilibrium to be reached and are less sensitive to competition with physiological substrates and/or ligands
(for example, orlistat binds irreversibly to the active site serine of pancreatic lipase, azacitidine irreversibly binds to DNA
methyltransferases and candesartan has a slow dissociation rate from the angiotensin II receptor)
1720,35,63,69
.
Conformational mechanisms
For conformational mechanisms, binding of the drug to the target involves a conformational change in the target that
couples drug binding to a response (for example, sirolimus binds to the peptidylprolyl isomerase FKBP12, which stabilizes
a conformation that subsequently inhibits the kinase activity of mammalian target of rapamycin; and fulvestrant induces a
conformation of the oestrogen receptor that is subsequently degraded)
811,47,70
.
Noncompetitive inhibition and/or antagonism. This is a form of MMOA in which the drug binds to a site on the target that
is distinct from the physiological substrate- and/or ligand-binding site that results in an inhibition of the response (for
example, caspofungin is a noncompetitive inhibitor of 1,3--d-glucan synthase owing to the observation that its IC
50

(half-maximal inhibitory concentration) is not influenced by substrate concentrations)
68,71
.
Uncompetitive inhibition and/or antagonism. An uncompetitive MMOA is contingent on prior activation of the target by
the physiological effector (the substrate or the ligand). This means that the same amount of drug blocks higher concentrations
of the physiological effector to a greater degree than lower concentrations. For example, memantine is an uncompetitive
antagonist that binds only to the activated form of the NMDA receptor. The potency of the inhibition of the NMDA receptor by
memantine increases at higher concentrations of glutamate (the physiological ligand)
22,23,68
.
Full agonism. Maximal efficacy is produced following drug binding to the receptor and subsequent receptor activation (for
example, ramelteon mimics the activity of melatonin for the melatonin receptor through binding at the orthosteric site
with efficient coupling to activate specific signalling pathways)
72,73
.
Partial agonism. This is a form of MMOA in which only partial efficacy is produced following drug binding to the orthosteric
site on the receptor (for example, aripiprazole is a partial agonist of the dopamine D2 receptor and varenciline is a partial
agonist of the nicotinic acetylcholine receptors )
73,7476
.
Allosteric modulator. This mechanism involves regulation of the biological activity of the target by binding of a drug at a
site other than the binding site for the endogenous substrate and/or ligand (allosteric site) (for example, cinacalcet is an
allosteric modulator of the calcium receptor by binding to the allosteric site)
29,73
.
Redox mechanisms
Redox is short for reductionoxidation reactions in which the pharmacological response to the drug is a consequence of
electron transfer between the drug and a physiological target. For example, generation of hydroxyl radicals by verteporfin
is thought to contribute to its ability to damage cells, and the antiprotozoal activity of nitazoxanide is believed to be due
in part to interference with the pyruvateferredoxin oxidoreductase enzyme-dependent electron transfer reaction, which
is essential to anaerobic energy metabolism
7779
.
ANALYSI S
nrd_3480_jul11.indd 511 22/06/2011 10:16
different conformations presumably change the affinity
of the receptor for the interacting co-activators and co-
repressors. The change in co-repressor affinity alters
the composition of the distinct cellular co-regulatory
complexes that modulate the functional transcriptional
activity
14
.

The MMOA can also differentiate similar drugs with
respect to their therapeutic indications. At the structural
level, aspirin is an irreversible inhibitor of cyclooxyge-
nases, whereas ibuprofen and naproxen are reversible
inhibitors. All three molecules bind to cyclooxygenase
enzymes at the same substrate binding site. However,
the irreversible MMOA of aspirin differentiates its
functional use as an antiplatelet drug from the revers-
ible inhibitors, because this MMOA translates into a
long-lasting action of aspirin in platelets, as platelets do
not have the capacity to resynthesize new enzymes
15,16
.
There are many different biochemical features of
an MMOA through which molecular interactions can
contribute to a specific functional response. These
include residence time
10,1719
, irreversible binding
20
,
transient binding
21,22
, and uncompetitive
22,23
and non-
competitive
10
inhibition mechanisms (BOX 1). It has
been proposed that drugs should be activated by the
pathological state that they are intended to inhibit
22,23
.
Allosteric inhibition and activation are important for
the pharmacological modulation of many receptors and
channels
24,25
. Voltage- or frequency-dependent channel
blockade can also influence a selective pharmacological
response
26,27
. Given the importance of the MMOA to the
therapeutic effects of NMEs, we consider it further in
the following sections.
Discovery of first-in-class medicines
NMEs that were discovered through phenotypic screen-
ing. The 28 first-in-class small-molecule NMEs that
were discovered in phenotypic screens either came from
intentional targeting of a specific phenotype (25NMEs)
or through serendipity (3NMEs) (FIG.1). The inten-
tional approaches were based on assays that measured
a specific physiological phenomenon, with little under-
standing of the MMOA. In many cases, the newly dis-
covered molecules were subsequently used to identify
MMOAs for the physiological phenomena. For example,
the oxazolidinone antibiotics (such as linezolid) were
initially discovered as inhibitors of Gram-positive bac-
teria but were subsequently shown to be protein synthe-
sis inhibitors that target an early step in the binding of
N-formylmethionyl-tRNA to the ribosome
28
. This is also
illustrated by the calcium receptor allosteric activator cin-
acalcet
29
, the sterol transporter inhibitor ezetimibe
30
and
the N-type calcium channel blocker ziconotide
31
; these
drugs were initially discovered using phenotypic assays.
The majority of discoveries focused on using specific
chemical classes in which prior knowledge contributed
to matching the chemical class with the phenotype
for example, screening nucleoside analogues as poten-
tial anticancer and antiviral agents. Random library
screening was also successful for ezetimibe, linezolid,
pemirolast, retapamulin, rufinamide and sirolimus. An
additional approach was to use phenotypic screening
to identify new MMOAs for established targets, which
led to the discovery of the partial agonists aripiprazole
and varenicline, and the full antagonist fulvestrant (see
Supplementary information S2 (box) for details). It is
Table 1 cont. | First-in-class small-molecule new molecular entities approved by the FDA: 19992008
Imatinib (Gleevec; Novartis) Cancer Enzyme Stabilizes inactive conformation 49
Maraviroc (Celsentri; Pfizer) Infectious disease Receptor Conformational and/or allosteric 134137
Mifepristone (Mifeprex; Aventis
Pharma)
Reproductive Receptor Conformational antagonist 138141
Orlistat (Xenical; Roche) Metabolic Enzyme Irreversible 35,142
Raltegravir (Isentress; Merck) Infectious disease Enzyme Traps conformational state 39,40,143,
144
Ramelteon (Rozerem; Takeda
Pharmaceuticals)
CNS Receptor Equilibrium binding 72,145
Sitagliptin (Januvia; Merck) Metabolic Enzyme Equilibrium binding 33,146
Sorafenib (Nexavar; Bayer) Cancer Enzyme Conformation state-specific inhibition 44
Sunitinib (Sutent; Pfizer) Cancer Enzyme Conformation state-specific inhibition 147150
Zanamivir
(Relenza; GlaxoSmithKline)
Infectious disease Enzyme Equilibrium binding 34,151
Discovered based on natural substrate or natural substance
Acamprosate (Campral; Merck) CNS Ion channel Conformational channel modulator 152
Aminolevulinic acid (Levulan; Berlex) Dermatology NA (photosensitizer) Redox 153,154
Fondaparinux (Arixtra; Sanofi) Cardiovascular Enzyme Irreversible 155157
Sapropterin (Kuvan; BioMarin) Rare diseases Enzyme Cofactor 158161
Verteporfin (Visudyne; QLT) Ocular NA (photoreaction) Redox 77,162
CNS, central nervous system; FDA, US Food and Drug Administration; NA, not applicable.
ANALYSI S
nrd_3480_jul11.indd 510 22/06/2011 10:16
0CVWTG4GXKGYU&TWI&KUEQXGT[
2JGPQV[RKE
UETGGPKPI
lnonionul urqoinq
ol socic bonoyo
Abuuco
Aqulsiduso-
Aloluco
Alomuzumub
Alqlucosiduso ullu
Anulinru
8ovucizumub
Couximub
Doniloulin
Drorocoqin-
Lculizumub
Llulizumub
Lnluvirido
Lxonuido
Gulsulluso
Gomuzumub
ldursulluso
Luroniduso
Nuulizumub
Omulizumub
Pulilormin
Poqvisomun
Prumlinido
Pusburicuso
Pomilosim
Oimizod MMOA
subsoquonly
idoniod
C
6CTIGVDCUGFUETGGPKPI D
F $KQNQIKEU

Duomycin
Lzoimibo
Linozolid
Nuoqlinido
Pomirolus
Punumido
Scrooninq ol
rundom
comound
librury
Momunino
Sinocuocbins
Vorinosu
Azuciidino
Cusolunqin
Cilosuzol
Cinuculco
Docosunol
Lovoirucoum
Lubirosono
Miqlusu
Nolurubino
Niuzoxunido
Niisinono
Punoluzino
Pouumulin
Sirolimus
Ziconoido
Zonisumido
Scrooninq ol comound-socic librurios
busod on siqnicun rior lnovlodqo
ol comound roorios
Ariiruzolo
lulvosrun
Vuroniclino
Knovn urqo:
soolinq imrovod
MMOA
Goinib
lmuinib
Muruviroc
Puloqruvir
Sorulonib
Suniinib
Alisliron
Aroiun
8orozomib
8osonun
Conivuun
Llrombouq
Orlisu
Siuqliin
Zunumivir
Milorisono
Pumoloon
Modiod nuurul
urqo liqund
Acumrosuo
Aminolovulinic ucid
londuurinux
Suroorin
Voroorn
E 5[PVJGVKEPCVWTCN
UWDUVCPEGU
Sorondiious
discovorios
omalizumab, pegvisomant and natalizumab); and deliv-
ering other compounds or proteins (denileukin diftitox
and gemtuzumab). Thus, the majority of these biologics
function by interfering with a molecular activity and, as
mentioned above, all of these biologics can be considered
to have been discovered using a target-based approach.
Both first-in-class small molecule NMEs and bio-
logics were approved for two targets: EGFR kinase (the
small-molecule EGFR kinase inhibitor gefitinib and
the EGFR-specific monoclonal antibody cetuximab)
and TPO (the small-molecule TPO receptor agonist
eltrombopag and the peptibody TPO receptor agonist
romiplostim). Three first-in-class medicines also act by
inhibiting vascular endothelial growth factor (VEGF)
signalling: the VEGF-specific monoclonal antibody
bevacizumab, and the small-molecule VEGF receptor
kinase inhibitors sunitinib, which also inhibits KIT, and
sorafenib, which was originally discovered on the basis
of its inhibition of RAF kinase
44
.
Strategies according to disease area. Evaluation of the
discovery strategy by disease area showed that a pheno-
typic approach was the most successful for central nerv-
ous system disorders and infectious diseases, whereas
target-based approaches were most successful in can-
cer, infectious diseases and metabolic diseases (TABLE2).
Biologics accounted for most of the new medicines that
act by modulating the immune system and 50% of the
new medicines forcancer.
Discovery of follower drugs
There were 164 follower drugs, out of which 83 (51%)
were discovered via target-based approaches, 30 (18%)
via phenotypic assays and 31 (19%) were biologics (FIG.2)
(Supplementary information S1 (table)). Seven (4%) of
the follower drugs were prodrugs or combinations of
previously approved medicines. Considering NMEs
alone, target-based approaches accounted for 62% (83
out of 133) of the small-molecule NMEs. The ratio of
NMEs from target-based approaches to those from phe-
notypic screening increased during the final 4years of
the analysis (FIG.3b).
Molecular mechanism of action
The majority of small-molecule first-in-class NMEs had
MMOAs that involved inhibiting the activity of enzymes or
modulating receptors (FIG.4). This trend is consistent with
the findings of Imming and colleagues
4
in their analysis
of the nature and number of all drug targets. The phar-
macological responses were often achieved by binding to
the target protein to elicit a positive or negative response.
For the first-in-class NMEs and biologics, many
different biochemical mechanisms mediated the drug
response at the target (BOX1). These included revers-
ible, irreversible and slow binding kinetics; competi-
tive, uncompetitive and noncompetitive interactions
between physiological substrates/ligands and drugs; as
well as inhibition, activation, agonism, partial agonism,
allosteric activation and induced degradation.
Illustrative examples in which stimulation of a bio-
logical response was achieved included: exenatide,
which mimics a natural peptide (glucagon-like peptide1
(GLP1)) but is resistant to degradation by the protease
DPP4 (REF.45); sitagliptin, which prevents degradation
of endogenous GLP1 by inhibiting DPP4 (REF.33); and
cinacalcet, which is an allosteric activator of the calcium-
sensing receptor
29
.
Illustrative examples in which inhibition or antago-
nism of a biological response was achieved included:
aprepitant, which is a competitive antagonist of the
neurokinin-1 receptor
46
; orlistat, which is an irreversible
inhibitor of lipase enzymes
35
; fulvestrant, which induces
Figure 1 | Discovery strategies used to identify first-in-class medicines. The
strategies that were used were categorized as being based on phenotypic screening (a),
target-based strategies (b), synthetic versions of natural substances or very close
derivatives (c) and biologics (d). Phenotypic strategies were further subdivided into
intentional screening with random compound libraries or compound-specific libraries,
optimization for molecular mechanism of action (MMOA) and serendipitous discoveries.
Drugs that were identified through target-based screening that involved optimization of
a natural ligand or identification of the optimal MMOA are highlighted. *Drugs that are
derived from natural substances.
These medicines have been withdrawn from the market.
Although enfuvirtide and pegvisomant were approved as new molecular entities, for the
purpose of this analysis they have been treated as biologics, given that they are both much
larger than typical small-molecule drugs (see Supplementary information S2 (box)).
ANALYSI S
nrd_3480_jul11.indd 513 22/06/2011 10:16
acamprosate (FIG. 1c). Additionally, in some cases,
natural substances provided starting points for small-
molecule phenotypic screening (10 NMEs (FIG. 1a)) and
target-based discovery (3 NMEs (FIG. 1b)). In total, 18
out of the 50 (36%) first-in-class small-molecule NMEs
originated from natural substances. These numbers are
consistent with those reported by Newman and Cragg
32

for the percentage of all medicines derived from natu-
ral products, and the supposition that libraries that are
derived from natural substances provide good chemi-
cal starting points for optimization. For example, two
NMEs that were discovered using a target-specific
strategy ramelteon, which targets melatonin recep-
tors, and mifepristone, which is a progesterone receptor
modulator were derived from the modification of
natural ligands.
Target-based approaches. Target-based approaches
led to the discovery of 17 of the 50 first-in-class small-
molecule NMEs. Various approaches contributed to
these discoveries, and they are illustrated by the fol-
lowing examples. Sitagliptin, an inhibitor of the pro-
tease dipeptidyl peptidase 4 (DPP4), was discovered
in an iterative discovery approach that was aimed at
optimizing metabolic properties while retaining effi-
cacy
33
. A computer-assisted drug design strategy that
was based on the crystal structure of the influenza viral
neuraminidase led to the identification of zanamivir
34
.
A target-directed screening of microbial broths from
soil organisms resulted in the discovery of a very potent,
selective and irreversible inhibitor of pancreatic lipases,
which was named lipstatin (orlistat)
35
. Eltrombopag was
identified by screening small-molecule libraries for the
ability to activate a reporter molecule in thrombopoietin
(TPO)-dependent cell lines. Lead compounds were ini-
tially identified and then optimized for their biological
effect and pharmaceutical properties
36
. In a programme
that was aimed at discovering non-peptide endothelin
receptor antagonists, a class of substituted arylsulphon-
amidopyrimidines was identified in a chemical com-
pound library, which led to the discovery of bosentan
37
.
However, knowledge of the targets did not necessarily
lead to an easy path to discovery. For example, although
renin had been a clear target for the treatment of hyper-
tension for decades, the development of orally active
renin inhibitors, which culminated in the discovery of
the NME aliskiren
38
, was a major challenge.
The development of six of the NMEs that were dis-
covered by target-based approaches involved subsequent
identification of their effective MMOA at the target that
was selected for the initial screening strategies. The
kinase inhibitors gefitinib, imatinib, sorafenib and suni-
tinib block kinase activation; the HIV integrase inhibi-
tor raltegravir traps an intermediate complex between
the enzyme and nucleic acid; and maraviroc is an allos-
teric antagonist of the the CC chemokine receptor type
5. These inhibitors represent successes of the target-
based strategy, but they also highlight that the optimal
MMOA at the target may not be apparent at the time of
initiating the discovery strategy. For example, the HIV1
integrase inhibitor raltegravir was only discovered after
several MMOAs had been investigated using different
assay formats
39,40
. The diketo acids that led to the dis-
covery of raltegravir were eventually found to block the
strand transfer reaction, and this MMOA provided good
invivo efficacy. The importance of the assay format in
the identification of compounds with effective MMOAs
at a chosen target is also illustrated by the discovery of
gefitinib, which is thought to act by sequestering the
EGFR and its ligand into inactive receptorligand com-
plexes
41
. Screening for activity in A431 vulval squamous
carcinoma cells was the assay format that led to the iden-
tification of gefitinib and its MMOA
42
.
The neurokinin-1 receptor antagonist aprepitant and
the proteasome inhibitor bortezomib were originally dis-
covered with a view to targeting different indications to
those that they were first approved for (Supplementary
information S2 (box)). Repositioning was also involved
for three of the NMEs that were discovered through
phenotypic assays: miglustat, azacitidine and nitisinone
(Supplementary information S2 (box)).
Biologics. Biologics that were approved under biolog-
ics license applications and large peptide molecules
that were approved as NMEs (for example, enfuvirtide
and pegvisomant) accounted for 25 (33%) out of the
75 first-in-class medicines (FIG. 1d). The biologics were
further categorized according to their pharmacological
action as described by Leader, Baca and Golan
43
. The
pharmacological actions of these biologics included
enzyme replacement (agalsidase-, alglucosidase alfa,
galsulfase, idursulfase and laronidase), augmenting
existing pathways (drotrecogin-, exenatide, palifermin,
pramlintide and romiplostim), providing a novel func-
tion (rasburicase), interfering with a molecular activity
(alemtuzumab, abatacept, anakinra, alefacept, bevaci-
zumab, cetuximab, eculizumab, efalizumab, enfuvirtide,

Box 2 | Biochemical efficiency
The dose of a drug required to achieve the desired physiological response depends on
its biochemical efficiency
10,11
. This is defined as binding affinity/functional response,
which is equivalent to K
i
/EC
50
(effector concentration for half-maximal response)

. Good
biochemical efficiency enables efficacy at lower drug concentrations and increases the
therapeutic index. It is a property of many approved medicines
10,11
.
There are many factors that can influence the shift in doseresponse curves between
binding and functional assays, including:
Pharmacokinetics and ADME (absorption, distribution, metabolism and excretion)
properties
Assay relevance (is the functional assay appropriate for the target? Are the assays
technically accurate?)
The involvement of the target in the functional readout and biology
The molecular mechanism of action (MMOA)
Although all of these factors can and do contribute to the relationship between
binding affinity and the functional response, the role of the MMOA is not always
considered. The concept of biochemical efficiency was introduced to quantify this
possibility
10,11
. When biochemical efficiency is used as a measure of an optimal MMOA,
it is important that the other mitigating factors are eliminated. For example, when
evaluating biochemical efficiency, the assays must be run in the absence of serum
(or plasma) to eliminate the shift in IC
50
(half-maximal inhibitory concentration) owing
to serum protein binding.
ANALYSI S
nrd_3480_jul11.indd 512 22/06/2011 10:16
2JGPQV[RKE
UETGGPKPI
lnonionul urqoinq
ol socic bonoyo
Abuuco
Aqulsiduso-
Aloluco
Alomuzumub
Alqlucosiduso ullu
Anulinru
8ovucizumub
Couximub
Doniloulin
Drorocoqin-
Lculizumub
Llulizumub
Lnluvirido
Lxonuido
Gulsulluso
Gomuzumub
ldursulluso
Luroniduso
Nuulizumub
Omulizumub
Pulilormin
Poqvisomun
Prumlinido
Pusburicuso
Pomilosim
Oimizod MMOA
subsoquonly
idoniod
C
6CTIGVDCUGFUETGGPKPI D
F $KQNQIKEU

Duomycin
Lzoimibo
Linozolid
Nuoqlinido
Pomirolus
Punumido
Scrooninq ol
rundom
comound
librury
Momunino
Sinocuocbins
Vorinosu
Azuciidino
Cusolunqin
Cilosuzol
Cinuculco
Docosunol
Lovoirucoum
Lubirosono
Miqlusu
Nolurubino
Niuzoxunido
Niisinono
Punoluzino
Pouumulin
Sirolimus
Ziconoido
Zonisumido
Scrooninq ol comound-socic librurios
busod on siqnicun rior lnovlodqo
ol comound roorios
Ariiruzolo
lulvosrun
Vuroniclino
Knovn urqo:
soolinq imrovod
MMOA
Goinib
lmuinib
Muruviroc
Puloqruvir
Sorulonib
Suniinib
Alisliron
Aroiun
8orozomib
8osonun
Conivuun
Llrombouq
Orlisu
Siuqliin
Zunumivir
Milorisono
Pumoloon
Modiod nuurul
urqo liqund
Acumrosuo
Aminolovulinic ucid
londuurinux
Suroorin
Voroorn
E 5[PVJGVKEPCVWTCN
UWDUVCPEGU
Sorondiious
discovorios
omalizumab, pegvisomant and natalizumab); and deliv-
ering other compounds or proteins (denileukin diftitox
and gemtuzumab). Thus, the majority of these biologics
function by interfering with a molecular activity and, as
mentioned above, all of these biologics can be considered
to have been discovered using a target-based approach.
Both first-in-class small molecule NMEs and bio-
logics were approved for two targets: EGFR kinase (the
small-molecule EGFR kinase inhibitor gefitinib and
the EGFR-specific monoclonal antibody cetuximab)
and TPO (the small-molecule TPO receptor agonist
eltrombopag and the peptibody TPO receptor agonist
romiplostim). Three first-in-class medicines also act by
inhibiting vascular endothelial growth factor (VEGF)
signalling: the VEGF-specific monoclonal antibody
bevacizumab, and the small-molecule VEGF receptor
kinase inhibitors sunitinib, which also inhibits KIT, and
sorafenib, which was originally discovered on the basis
of its inhibition of RAF kinase
44
.
Strategies according to disease area. Evaluation of the
discovery strategy by disease area showed that a pheno-
typic approach was the most successful for central nerv-
ous system disorders and infectious diseases, whereas
target-based approaches were most successful in can-
cer, infectious diseases and metabolic diseases (TABLE2).
Biologics accounted for most of the new medicines that
act by modulating the immune system and 50% of the
new medicines forcancer.
Discovery of follower drugs
There were 164 follower drugs, out of which 83 (51%)
were discovered via target-based approaches, 30 (18%)
via phenotypic assays and 31 (19%) were biologics (FIG.2)
(Supplementary information S1 (table)). Seven (4%) of
the follower drugs were prodrugs or combinations of
previously approved medicines. Considering NMEs
alone, target-based approaches accounted for 62% (83
out of 133) of the small-molecule NMEs. The ratio of
NMEs from target-based approaches to those from phe-
notypic screening increased during the final 4years of
the analysis (FIG.3b).
Molecular mechanism of action
The majority of small-molecule first-in-class NMEs had
MMOAs that involved inhibiting the activity of enzymes or
modulating receptors (FIG.4). This trend is consistent with
the findings of Imming and colleagues
4
in their analysis
of the nature and number of all drug targets. The phar-
macological responses were often achieved by binding to
the target protein to elicit a positive or negative response.
For the first-in-class NMEs and biologics, many
different biochemical mechanisms mediated the drug
response at the target (BOX1). These included revers-
ible, irreversible and slow binding kinetics; competi-
tive, uncompetitive and noncompetitive interactions
between physiological substrates/ligands and drugs; as
well as inhibition, activation, agonism, partial agonism,
allosteric activation and induced degradation.
Illustrative examples in which stimulation of a bio-
logical response was achieved included: exenatide,
which mimics a natural peptide (glucagon-like peptide1
(GLP1)) but is resistant to degradation by the protease
DPP4 (REF.45); sitagliptin, which prevents degradation
of endogenous GLP1 by inhibiting DPP4 (REF.33); and
cinacalcet, which is an allosteric activator of the calcium-
sensing receptor
29
.
Illustrative examples in which inhibition or antago-
nism of a biological response was achieved included:
aprepitant, which is a competitive antagonist of the
neurokinin-1 receptor
46
; orlistat, which is an irreversible
inhibitor of lipase enzymes
35
; fulvestrant, which induces
Figure 1 | Discovery strategies used to identify first-in-class medicines. The
strategies that were used were categorized as being based on phenotypic screening (a),
target-based strategies (b), synthetic versions of natural substances or very close
derivatives (c) and biologics (d). Phenotypic strategies were further subdivided into
intentional screening with random compound libraries or compound-specific libraries,
optimization for molecular mechanism of action (MMOA) and serendipitous discoveries.
Drugs that were identified through target-based screening that involved optimization of
a natural ligand or identification of the optimal MMOA are highlighted. *Drugs that are
derived from natural substances.
These medicines have been withdrawn from the market.
Although enfuvirtide and pegvisomant were approved as new molecular entities, for the
purpose of this analysis they have been treated as biologics, given that they are both much
larger than typical small-molecule drugs (see Supplementary information S2 (box)).
ANALYSI S
nrd_3480_jul11.indd 513 22/06/2011 10:16
acamprosate (FIG. 1c). Additionally, in some cases,
natural substances provided starting points for small-
molecule phenotypic screening (10 NMEs (FIG. 1a)) and
target-based discovery (3 NMEs (FIG. 1b)). In total, 18
out of the 50 (36%) first-in-class small-molecule NMEs
originated from natural substances. These numbers are
consistent with those reported by Newman and Cragg
32

for the percentage of all medicines derived from natu-
ral products, and the supposition that libraries that are
derived from natural substances provide good chemi-
cal starting points for optimization. For example, two
NMEs that were discovered using a target-specific
strategy ramelteon, which targets melatonin recep-
tors, and mifepristone, which is a progesterone receptor
modulator were derived from the modification of
natural ligands.
Target-based approaches. Target-based approaches
led to the discovery of 17 of the 50 first-in-class small-
molecule NMEs. Various approaches contributed to
these discoveries, and they are illustrated by the fol-
lowing examples. Sitagliptin, an inhibitor of the pro-
tease dipeptidyl peptidase 4 (DPP4), was discovered
in an iterative discovery approach that was aimed at
optimizing metabolic properties while retaining effi-
cacy
33
. A computer-assisted drug design strategy that
was based on the crystal structure of the influenza viral
neuraminidase led to the identification of zanamivir
34
.
A target-directed screening of microbial broths from
soil organisms resulted in the discovery of a very potent,
selective and irreversible inhibitor of pancreatic lipases,
which was named lipstatin (orlistat)
35
. Eltrombopag was
identified by screening small-molecule libraries for the
ability to activate a reporter molecule in thrombopoietin
(TPO)-dependent cell lines. Lead compounds were ini-
tially identified and then optimized for their biological
effect and pharmaceutical properties
36
. In a programme
that was aimed at discovering non-peptide endothelin
receptor antagonists, a class of substituted arylsulphon-
amidopyrimidines was identified in a chemical com-
pound library, which led to the discovery of bosentan
37
.
However, knowledge of the targets did not necessarily
lead to an easy path to discovery. For example, although
renin had been a clear target for the treatment of hyper-
tension for decades, the development of orally active
renin inhibitors, which culminated in the discovery of
the NME aliskiren
38
, was a major challenge.
The development of six of the NMEs that were dis-
covered by target-based approaches involved subsequent
identification of their effective MMOA at the target that
was selected for the initial screening strategies. The
kinase inhibitors gefitinib, imatinib, sorafenib and suni-
tinib block kinase activation; the HIV integrase inhibi-
tor raltegravir traps an intermediate complex between
the enzyme and nucleic acid; and maraviroc is an allos-
teric antagonist of the the CC chemokine receptor type
5. These inhibitors represent successes of the target-
based strategy, but they also highlight that the optimal
MMOA at the target may not be apparent at the time of
initiating the discovery strategy. For example, the HIV1
integrase inhibitor raltegravir was only discovered after
several MMOAs had been investigated using different
assay formats
39,40
. The diketo acids that led to the dis-
covery of raltegravir were eventually found to block the
strand transfer reaction, and this MMOA provided good
invivo efficacy. The importance of the assay format in
the identification of compounds with effective MMOAs
at a chosen target is also illustrated by the discovery of
gefitinib, which is thought to act by sequestering the
EGFR and its ligand into inactive receptorligand com-
plexes
41
. Screening for activity in A431 vulval squamous
carcinoma cells was the assay format that led to the iden-
tification of gefitinib and its MMOA
42
.
The neurokinin-1 receptor antagonist aprepitant and
the proteasome inhibitor bortezomib were originally dis-
covered with a view to targeting different indications to
those that they were first approved for (Supplementary
information S2 (box)). Repositioning was also involved
for three of the NMEs that were discovered through
phenotypic assays: miglustat, azacitidine and nitisinone
(Supplementary information S2 (box)).
Biologics. Biologics that were approved under biolog-
ics license applications and large peptide molecules
that were approved as NMEs (for example, enfuvirtide
and pegvisomant) accounted for 25 (33%) out of the
75 first-in-class medicines (FIG. 1d). The biologics were
further categorized according to their pharmacological
action as described by Leader, Baca and Golan
43
. The
pharmacological actions of these biologics included
enzyme replacement (agalsidase-, alglucosidase alfa,
galsulfase, idursulfase and laronidase), augmenting
existing pathways (drotrecogin-, exenatide, palifermin,
pramlintide and romiplostim), providing a novel func-
tion (rasburicase), interfering with a molecular activity
(alemtuzumab, abatacept, anakinra, alefacept, bevaci-
zumab, cetuximab, eculizumab, efalizumab, enfuvirtide,

Box 2 | Biochemical efficiency
The dose of a drug required to achieve the desired physiological response depends on
its biochemical efficiency
10,11
. This is defined as binding affinity/functional response,
which is equivalent to K
i
/EC
50
(effector concentration for half-maximal response)

. Good
biochemical efficiency enables efficacy at lower drug concentrations and increases the
therapeutic index. It is a property of many approved medicines
10,11
.
There are many factors that can influence the shift in doseresponse curves between
binding and functional assays, including:
Pharmacokinetics and ADME (absorption, distribution, metabolism and excretion)
properties
Assay relevance (is the functional assay appropriate for the target? Are the assays
technically accurate?)
The involvement of the target in the functional readout and biology
The molecular mechanism of action (MMOA)
Although all of these factors can and do contribute to the relationship between
binding affinity and the functional response, the role of the MMOA is not always
considered. The concept of biochemical efficiency was introduced to quantify this
possibility
10,11
. When biochemical efficiency is used as a measure of an optimal MMOA,
it is important that the other mitigating factors are eliminated. For example, when
evaluating biochemical efficiency, the assays must be run in the absence of serum
(or plasma) to eliminate the shift in IC
50
(half-maximal inhibitory concentration) owing
to serum protein binding.
ANALYSI S
nrd_3480_jul11.indd 512 22/06/2011 10:16
30
C D
2S
20
N
u
m
b
o
r

o
l

N
M
L
s
N
u
m
b
o
r

o
l

N
M
L
s
1S
10
S
0
00
80
10
60
S0
40
30
20
10
0
Your
1000 2000 2001 2002 2003 2004 200S 2006 2001 2008
Your
1000 2000 2001 2002 2003 2004 200S 2006 2001 2008
Pbonoyic scrooninq
Turqo-busod scrooninq
8ioloqics
this proposal is that the MMOA is a key factor for the
success of all approaches, but is addressed in different
ways and at different points in the various approaches.
In the more common target-based approach, drug
discovery is generally hypothesis-driven, and there are
at least three hypotheses that must be correct to result
in a new drug. The first hypothesis, which also applies
to other discovery approaches, is that activity in the
preclinical screens that are used to select a drug candi-
date will translate effectively into clinically meaningful
activity in patients. The other two hypotheses are that
the target that is selected is important in human disease
and that the MMOA of drug candidates at the target in
question is one that is capable of achieving the desired
biological response. Successful target-based discov-
ery of first-in-class drugs with tolerable safety profiles
requires the time and resources to investigate all three
hypotheses. In particular, the importance of hypoth-
esis testing to identify an appropriate MMOA may be
an underappreciated challenge that if neglected
could contribute to increased attrition rates for such
approaches. In other words, it is clearly difficult to
rationally identify the specific molecular interactions
from all of the potential dynamic molecular interac-
tions that will contribute to an optimal MMOA. Thus,
the key biochemical nuances that are important for the
translation of the molecular interaction (between a drug
and the target) to an optimal pharmacological response
could be missed with target-based approaches.
By contrast, in the case of phenotypic-based screening
approaches, assuming that a screening assay that trans-
lates effectively to human disease is available or can be
identified, a potential key advantage of this approach over
target-based approaches is that there is no preconceived
idea of the MMOA and target hypothesis. This could
considerably aid the identification of molecules with
appropriate targets (and possibly multiple targets) and
MMOAs, which might be less likely to emerge rapidly, if
at all, from pursuing a focused target-based hypothesis.
However, two limitations of phenotypic-based screening
approaches should also be noted. First, it will often be
necessary to characterize the MMOA of active molecules
that are identified in phenotypic screens to aid the opti-
mization of a drug candidate, but substantial progress has
been made in approaches to achieve this for example,
approaches based on RNA interference
54,55
. Second, phe-
notypic assays are often lower in throughput than stand-
ard target-based assays, although considerable progress
has also been made in recent years to automate such
assays and increase their throughput
5658
.
Finally, as has often been noted in reviews of the role
of natural products in drug discovery
32,59
, discovery
strategies that are based on natural substances have an
inherent advantage: the biology, target and MMOA are
often likely to be have been optimized already through
evolution, and so modifying such substances can be a
fruitful approach. Similarly, some of the biologics that
have been approved are harnessing endogenous mecha-
nisms in a rational way for example, by providing a
natural protein that is reduced in a given disease state,
as is the case for enzyme replacement therapies for
lyosomal storage disorders. In other cases though, it is
apparent that the precise MMOA of biologics might also
be important in their biological effects, as illustrated by
the differences in the properties of two monoclonal
antibodies that target CD20 on Bcells
60
rituximab
and ofatumumab although neither of these were
approved in the 10-year period we studied. Telling et
al.
60
conclude that the recognition of a novel epitope
cooperates with a slow off-rate in determining the activ-
ity of CD20 monoclonal antibodies in the activation of
complement and the induction of tumour celllysis.
The importance of the MMOA is further supported
by the evolution of the MMOA within drug classes, from
the first-in-class molecule to the best-in-class molecule,
which is not widely appreciated. For example, in some
cases in which there is no mechanism-based toxicity, the
evolution of drugs in a given class towards the best-in-
class has been associated with slower dissociation rates
at the target. This has been observed with antihistamines
Figure 3 | Cumulative distribution of new drugs by discovery strategy. a | First-in-class drugs. A lag is not strongly
apparent in a comparison of the cumulative number of small-molecule new molecular entities (NMEs) that were
discovered from the different approaches during the period ana lysed. b | Follower drugs. For follower drugs, the ratio
of small-molecule NMEs discovered through target-based screening to those discovered through phenotypic screening
appears to increase in the second half of the time period.
ANALYSI S
nrd_3480_jul11.indd 515 22/06/2011 10:16
degradation of the oestrogen receptor
47
; bevacizumab,
which binds to VEGF, thereby preventing its interaction
with its cell surface receptors
48
; and imatinib, which
inhibits the BCRABL kinase by stabilizing its inac-
tive conformation
49
(see Supplementary information S2
(box)) for further details on these and other MMOAs).
Importantly, simple equilibrium binding at the target
was rarely sufficient for the translation of drug binding
to the target into a therapeutically useful response a
subtle aspect of drug action that is underappreciated.
These results are consistent with the previous conclu-
sion
10
that two components are important to the MMOA.
The first component is the initial mass action-dependent
interaction. The second component requires a coupled
biochemical event to create a transition away from mass-
action equilibrium. It is also consistent with the opinions
expressed by Imming and colleagues
4
in their analysis
of drug targets, in which they emphasized the need to
consider the dynamics of the drugtarget interactions,
because in situations in which the dynamic actions of the
drug substance stimulate, or inhibit, a biological process, it
is necessary to move away from the descriptions of single
proteins, receptors and so on and to view the entire signal
chain as the target.
The diversity of the MMOAs of the new drugs ana-
lysed in this article is not surprising. Physiological and
drug mechanisms provide numerous examples of how
diversity and complexity in the MMOA can provide
robust, selective and timely functional responses. For
example, nuclear receptor ligands can induce ligand-
specific structural conformations that can be uniquely
coupled to the physiological system to provide func-
tionally selective responses
14
. Such conformational
changes might not be detectable by X-ray crystallogra-
phy studies; indeed, this was recently demonstrated for
the
2
-adrenergic receptor there was no discernable
difference in the conformation of the receptor when it
was bound to an inverse agonist or an antagonist
50
. The
functions of many enzymes are also regulated by specific
structural changes. For example, receptor tyrosine kinase
activation requires conformational changes that are
facilitated by ligand binding
51
, and many proteases have
inhibitory domains that must be proteolytically cleaved
for enzyme activation
52
. Both kinetics and conformation
contribute to the specificity of high-fidelity nucleotide
incorporation by DNA polymerases. Kinetic analysis has
shown that the nucleotide substrate-induced structural
change has a key role in discriminating between cor-
rect and incorrect base pairs, by governing whether a
nucleotide will be retained and incorporated or rapidly
released
53
.
Discussion
A principal observation from this analysis is that the
majority of small-molecule first-in-class NMEs that were
discovered between 1999 and 2008 were first discovered
using phenotypic assays (FIG.2): 28 of the first-in-class
NMEs came from phenotypic screening approaches,
compared with 17 from target-based approaches. This is
despite the current focus of small-molecule drug discov-
ery on target-based approaches. A possible contributing
factor to this trend could have been a lag time between
the introduction of new technologies and strategies, and
their impact in terms of the number of approved first-
in-class NMEs derived from these approaches. However,
such a lag is not strongly apparent in a comparison of the
cumulative number of NMEs from the two approaches
during the period analysed (FIG.3a).
This observation, along with further analysis of the
MMOA of the first-in-class NMEs, leads us to propose
that a focus on target-based drug discovery, without
accounting sufficiently for the MMOA of small-mole-
cule first-in-class medicines, could be a technical reason
contributing to high attrition rates. Our reasoning for
Table 2 | Discovery of first-in-class NMEs by therapeutic area
Disease area Target-based
screening
Phenotypic
screening
Biologics
Infectious diseases 3 7 1
Immune 1 0 6
Cancer 5 3 8
Central nervous system 1 7 1
Metabolic 3 2 2
Cardiovascular 2 3 0
Gastrointestinal 1 1 1
Others 1 3 1
Rare diseases 0 2 5
NME, new molecular entity.
S0
40
30
P
o
r
c
o
n
u
q
o

o
l

N
M
L
s
28
Pbonoyic
scrooninq
11
2S
S
30
83
31
13
31
23
33
1
18
S1
10
8
20
10
0
lirs-in-cluss druq lollovor druq
Turqo-
busod
scrooninq
Modiod
nuurul
subsuncos
8ioloqics
Figure 2 | The distribution of new drugs discovered
between 1999 and 2008, according to the discovery
strategy. The graph illustrates the number of new molecular
entities (NMEs) in each category. Phenotypic screening was
the most successful approach for first-in-class drugs,
whereas target-based screening was the most successful for
follower drugs during the period of this analysis. The total
number of medicines that were discovered via phenotypic
assays was similar for first-in-class and follower drugs
28 and 30, respectively whereas the total number of
medicines that were discovered via target-based screening
was nearly five times higher for follower drugs versus
first-in-class drugs (83 to 17, respectively).
ANALYSI S
nrd_3480_jul11.indd 514 22/06/2011 10:16
30
C D
2S
20
N
u
m
b
o
r

o
l

N
M
L
s
N
u
m
b
o
r

o
l

N
M
L
s
1S
10
S
0
00
80
10
60
S0
40
30
20
10
0
Your
1000 2000 2001 2002 2003 2004 200S 2006 2001 2008
Your
1000 2000 2001 2002 2003 2004 200S 2006 2001 2008
Pbonoyic scrooninq
Turqo-busod scrooninq
8ioloqics
this proposal is that the MMOA is a key factor for the
success of all approaches, but is addressed in different
ways and at different points in the various approaches.
In the more common target-based approach, drug
discovery is generally hypothesis-driven, and there are
at least three hypotheses that must be correct to result
in a new drug. The first hypothesis, which also applies
to other discovery approaches, is that activity in the
preclinical screens that are used to select a drug candi-
date will translate effectively into clinically meaningful
activity in patients. The other two hypotheses are that
the target that is selected is important in human disease
and that the MMOA of drug candidates at the target in
question is one that is capable of achieving the desired
biological response. Successful target-based discov-
ery of first-in-class drugs with tolerable safety profiles
requires the time and resources to investigate all three
hypotheses. In particular, the importance of hypoth-
esis testing to identify an appropriate MMOA may be
an underappreciated challenge that if neglected
could contribute to increased attrition rates for such
approaches. In other words, it is clearly difficult to
rationally identify the specific molecular interactions
from all of the potential dynamic molecular interac-
tions that will contribute to an optimal MMOA. Thus,
the key biochemical nuances that are important for the
translation of the molecular interaction (between a drug
and the target) to an optimal pharmacological response
could be missed with target-based approaches.
By contrast, in the case of phenotypic-based screening
approaches, assuming that a screening assay that trans-
lates effectively to human disease is available or can be
identified, a potential key advantage of this approach over
target-based approaches is that there is no preconceived
idea of the MMOA and target hypothesis. This could
considerably aid the identification of molecules with
appropriate targets (and possibly multiple targets) and
MMOAs, which might be less likely to emerge rapidly, if
at all, from pursuing a focused target-based hypothesis.
However, two limitations of phenotypic-based screening
approaches should also be noted. First, it will often be
necessary to characterize the MMOA of active molecules
that are identified in phenotypic screens to aid the opti-
mization of a drug candidate, but substantial progress has
been made in approaches to achieve this for example,
approaches based on RNA interference
54,55
. Second, phe-
notypic assays are often lower in throughput than stand-
ard target-based assays, although considerable progress
has also been made in recent years to automate such
assays and increase their throughput
5658
.
Finally, as has often been noted in reviews of the role
of natural products in drug discovery
32,59
, discovery
strategies that are based on natural substances have an
inherent advantage: the biology, target and MMOA are
often likely to be have been optimized already through
evolution, and so modifying such substances can be a
fruitful approach. Similarly, some of the biologics that
have been approved are harnessing endogenous mecha-
nisms in a rational way for example, by providing a
natural protein that is reduced in a given disease state,
as is the case for enzyme replacement therapies for
lyosomal storage disorders. In other cases though, it is
apparent that the precise MMOA of biologics might also
be important in their biological effects, as illustrated by
the differences in the properties of two monoclonal
antibodies that target CD20 on Bcells
60
rituximab
and ofatumumab although neither of these were
approved in the 10-year period we studied. Telling et
al.
60
conclude that the recognition of a novel epitope
cooperates with a slow off-rate in determining the activ-
ity of CD20 monoclonal antibodies in the activation of
complement and the induction of tumour celllysis.
The importance of the MMOA is further supported
by the evolution of the MMOA within drug classes, from
the first-in-class molecule to the best-in-class molecule,
which is not widely appreciated. For example, in some
cases in which there is no mechanism-based toxicity, the
evolution of drugs in a given class towards the best-in-
class has been associated with slower dissociation rates
at the target. This has been observed with antihistamines
Figure 3 | Cumulative distribution of new drugs by discovery strategy. a | First-in-class drugs. A lag is not strongly
apparent in a comparison of the cumulative number of small-molecule new molecular entities (NMEs) that were
discovered from the different approaches during the period ana lysed. b | Follower drugs. For follower drugs, the ratio
of small-molecule NMEs discovered through target-based screening to those discovered through phenotypic screening
appears to increase in the second half of the time period.
ANALYSI S
nrd_3480_jul11.indd 515 22/06/2011 10:16
degradation of the oestrogen receptor
47
; bevacizumab,
which binds to VEGF, thereby preventing its interaction
with its cell surface receptors
48
; and imatinib, which
inhibits the BCRABL kinase by stabilizing its inac-
tive conformation
49
(see Supplementary information S2
(box)) for further details on these and other MMOAs).
Importantly, simple equilibrium binding at the target
was rarely sufficient for the translation of drug binding
to the target into a therapeutically useful response a
subtle aspect of drug action that is underappreciated.
These results are consistent with the previous conclu-
sion
10
that two components are important to the MMOA.
The first component is the initial mass action-dependent
interaction. The second component requires a coupled
biochemical event to create a transition away from mass-
action equilibrium. It is also consistent with the opinions
expressed by Imming and colleagues
4
in their analysis
of drug targets, in which they emphasized the need to
consider the dynamics of the drugtarget interactions,
because in situations in which the dynamic actions of the
drug substance stimulate, or inhibit, a biological process, it
is necessary to move away from the descriptions of single
proteins, receptors and so on and to view the entire signal
chain as the target.
The diversity of the MMOAs of the new drugs ana-
lysed in this article is not surprising. Physiological and
drug mechanisms provide numerous examples of how
diversity and complexity in the MMOA can provide
robust, selective and timely functional responses. For
example, nuclear receptor ligands can induce ligand-
specific structural conformations that can be uniquely
coupled to the physiological system to provide func-
tionally selective responses
14
. Such conformational
changes might not be detectable by X-ray crystallogra-
phy studies; indeed, this was recently demonstrated for
the
2
-adrenergic receptor there was no discernable
difference in the conformation of the receptor when it
was bound to an inverse agonist or an antagonist
50
. The
functions of many enzymes are also regulated by specific
structural changes. For example, receptor tyrosine kinase
activation requires conformational changes that are
facilitated by ligand binding
51
, and many proteases have
inhibitory domains that must be proteolytically cleaved
for enzyme activation
52
. Both kinetics and conformation
contribute to the specificity of high-fidelity nucleotide
incorporation by DNA polymerases. Kinetic analysis has
shown that the nucleotide substrate-induced structural
change has a key role in discriminating between cor-
rect and incorrect base pairs, by governing whether a
nucleotide will be retained and incorporated or rapidly
released
53
.
Discussion
A principal observation from this analysis is that the
majority of small-molecule first-in-class NMEs that were
discovered between 1999 and 2008 were first discovered
using phenotypic assays (FIG.2): 28 of the first-in-class
NMEs came from phenotypic screening approaches,
compared with 17 from target-based approaches. This is
despite the current focus of small-molecule drug discov-
ery on target-based approaches. A possible contributing
factor to this trend could have been a lag time between
the introduction of new technologies and strategies, and
their impact in terms of the number of approved first-
in-class NMEs derived from these approaches. However,
such a lag is not strongly apparent in a comparison of the
cumulative number of NMEs from the two approaches
during the period analysed (FIG.3a).
This observation, along with further analysis of the
MMOA of the first-in-class NMEs, leads us to propose
that a focus on target-based drug discovery, without
accounting sufficiently for the MMOA of small-mole-
cule first-in-class medicines, could be a technical reason
contributing to high attrition rates. Our reasoning for
Table 2 | Discovery of first-in-class NMEs by therapeutic area
Disease area Target-based
screening
Phenotypic
screening
Biologics
Infectious diseases 3 7 1
Immune 1 0 6
Cancer 5 3 8
Central nervous system 1 7 1
Metabolic 3 2 2
Cardiovascular 2 3 0
Gastrointestinal 1 1 1
Others 1 3 1
Rare diseases 0 2 5
NME, new molecular entity.
S0
40
30
P
o
r
c
o
n
u
q
o

o
l

N
M
L
s
28
Pbonoyic
scrooninq
11
2S
S
30
83
31
13
31
23
33
1
18
S1
10
8
20
10
0
lirs-in-cluss druq lollovor druq
Turqo-
busod
scrooninq
Modiod
nuurul
subsuncos
8ioloqics
Figure 2 | The distribution of new drugs discovered
between 1999 and 2008, according to the discovery
strategy. The graph illustrates the number of new molecular
entities (NMEs) in each category. Phenotypic screening was
the most successful approach for first-in-class drugs,
whereas target-based screening was the most successful for
follower drugs during the period of this analysis. The total
number of medicines that were discovered via phenotypic
assays was similar for first-in-class and follower drugs
28 and 30, respectively whereas the total number of
medicines that were discovered via target-based screening
was nearly five times higher for follower drugs versus
first-in-class drugs (83 to 17, respectively).
ANALYSI S
nrd_3480_jul11.indd 514 22/06/2011 10:16
belief by some that every target can provide the basis
for a drug. As such, research across the pharmaceutical
industry as well as academic institutions has increas-
ingly focused on targets, arguably at the expense of the
development of preclinical assays that translate more
effectively into clinical effects in patients with a specific
disease. In our analysis, we found that there are numerous
diverse MMOAs for approved new first-in-class drugs,
but drug discovery at present appears to be dominated by
a one size fits all approach, in which drugs are optimized
for binding affinity with less consideration for binding
kinetics and conformation. For optimal application of
target-based approaches, it is important to consider how
efficiently binding is coupled to the response (BOX2).
However, the molecular descriptors for the coupling fac-
tors may not be accurately captured by only consider-
ing binding affinity. Furthermore, an excessive focus on
affinity at a given target could lead to compromises being
made in pharmacokinetic properties that are critical for
the success of drugs, which has also been postulated to
be an underlying factor for current attritionrates
67
.
Reducing the impact of technical uncertainty on the
later, more costly stages of drug development through a
quick win/fast fail strategy has been proposed as a solu-
tion to the current problems with R&D productivity
2
.
However, this strategy does not address the key issues
that contribute to the greater technical uncertainty
and associated risk of failure. Our analysis leads us to
conclude that the identification of an optimal MMOA
has been a key factor contributing to the success of
phenotypic screening in the discovery of the first-in-
class NMEs in the 10-year period we studied. Thus, we
consider that technical risk and, consequently, overall
attrition in drug development could be decreased for
first-in-class drugs through the development and greater
use of translational phenotypic assays, and by considering
diverse MMOAs when using a target-based, hypothesis-
driven strategy.
1. Munos, B. Lessons for 60years of pharmaceutical
(2009).
2. Paul, S.M. etal. How to improve R.&D productivity:
the pharmaceutical industrys grand challenge. Nature
Rev. Drug Discov. 9, 203214 (2010).
3. Lindsay, M.A. Target discovery. Nature Rev. Drug
Discov. 2, 831838 (2003).
4. Imming, P., Sinning, C. & Meyer A. Drugs, their targets
and the nature and number of drug targets. Nature
Rev. Drug Discov. 5, 821834 (2006).
5. Overington, J.P., Al-Lazikani, B. & Hopkins, A.L. How
many drug targets are there? Nature Rev. Drug
Discov. 5, 993996(2006).
6. Williams, M. Systems and integrative biology as
alternative guises for pharmacology: prime time for an
iPharm concept? Biochem. Pharmacol. 70,
17071716 (2005).
7. Flordellis, C.S., Manolis, A.S., Paris, H. & Karabinis, A.
Rethinking target discovery in polygenic diseases.
Curr. Top. Med. Chem. 6, 17911798 (2006).
8. Urban, J.D. etal. Functional selectivity and classical
concepts of quantitative pharmacology J.Pharmacol.
Exp. Ther. 320, 113 (2007).
Formalizes the concept of functional selectivity,
whereby multiple unique ligands can bind to one
receptor to initiate different responses.
9. Kenakin, T. & Miller, L.J. Seven transmembrane
receptors as shapeshifting proteins: the impact of
allosteric modulation and functional selectivity on
drug discovery. Pharmacol. Rev. 62, 265304
(2010).
10. Swinney, D.C. Biochemical mechanisms of drug
action: what does it take for success? Nature Rev.
Drug Discov. 3, 801808 (2004).
Describes how the MMOA influences the
therapeutic index and utility of a medicine and
introduces biochemical efficiency as a metric to
quantify this influence.
11. Swinney, D.C. Biochemical mechanisms of new
molecular entities (NMEs) approved by United States
FDA during 20012004: mechanisms leading to
optimal efficacy and safety. Curr. Top. Med. Chem. 6,
461478 (2006).
12. Swinney, D.C. Applications of binding kinetics to drug
discovery: translation of binding mechanism to
clinically differentiated therapeutic responses. Pharm.
Med. 22, 2334 (2008).
13. Yun, C.H. etal. The T790M mutation in EGFR kinase
causes drug resistance by increasing the affinity for ATP.
Proc. Natl Acad. Sci. USA 105, 20702075 (2008).
Provides an illustration of how drug resistance
could be overcome through an understanding of
the MMOA.
14. Brzozowski, A.M. etal. Molecular basis of agonism
and antagonism in the oestrogen receptor. Nature
389, 753758 (1997).
Shows structurally how agonists and antagonists
bind at the same site but with different binding
modes that result in different responses.
15. Roth, G.J. & Majerus, P.W. The mechanism of the
effect of aspirin on human platelets. I. Acetylation of a
particulate fraction protein. J.Clin. Invest. 56,
624632 (1975).
16. Majerus, P.W., Broze, G.J. Jr, Miletich, J.P. & Tollefsen,
D.M. in Goodman & Gilmans The pharmacological
basis of therapeutics. (eds Hardman, J. G. & Limbird, L.
E.)1353 (McGraw-Hill, New York, 1996).
17. Copeland, R.A., Pompliano, D.L. & Meek, T.D.
Drug-target residence time and its implications for
lead optimization. Nature Rev. Drug Discov. 5,
730739 (2006).
18. Timmino, P.J. & Copeland, R.A. Residence time of
receptorligand complexes and its effect on biological
function. Biochemistry 47, 54815492 (2008).
19. Lu, H. & Tonge, P.J. Drug-target residence time:
critical information for lead optimization. Curr. Opin.
Chem. Biol. 14, 18 (2010).
20. Johnson, D.S., Weerapana, E. & Cravatt, B.F.
Strategies for discovering and derisking covalent,
irreversible enzyme inhibitors. Future Med. Chem. 2,
949964 (2010).
21. Ohlson, S. Designing transient binding drugs: a new
concept for drug discovery. Drug Discov. Today 13,
433439 (2008).
22. Lipton, S.A. Paradigm shift in neuroprotection by
NMDA receptor blockade: Memantine and beyond.
23. Lipton, S.A. Pathology activated therapeutics for
neuroprotection. Nature Rev. Neurosci. 8, 803808
(2007).
Describes the principle that drugs should be
activated by the pathological state that they are
intended to inhibit.
24. Changeux, J.P. Allosteric receptors: from electric
organ to cognition. Annu. Rev. Pharmacol. Toxicol. 50,
138 (2010).
25. Conn, J.P., Christopoulos, A. & Lindsley, C.W.
Allosteric modulators of GPCRs: a novel approach for
the treatment of CNS disorders. Nature Rev. Drug
Discov. 8, 4154 (2009).
26. Hanck, D.A. etal. Using lidocaine and benzocaine to
link sodium channel molecular conformations to state-
dependent antiarrhythic drug affinity. Circ. Res. 105,
492499 (2009).
27. Butterworth, J.F. & Strichartz, G.R. Molecular
mechanisms of local anesthesia: a review.
Anesthesiology 72, 711734 (1990).
28. Wilson, D.N. etal. The oxazolidinone antibiotics
perturb the ribosomal peptidyl-transferase center and
effect tRNA positioning. Proc. Natl Acad. Sci. USA
105, 1333913344 (2008).
29. Nemeth, E.F. Misconceptions about calcimimetics.
Ann. NYAcad. Sci. 1068, 471476 (2006).
Discusses lessons learned in the discovery of
cinacalcet, with emphasis on the importance of
using an understanding of physiology.
30. Salisbury, B.G. etal. Hypocholesterolemic activity of a
novel inhibitor of cholesterol absorption, SCH 48461.
Athlerosclerosis 115, 4563 (1995).
31. Valentino, D. etal. A selective N-type calcium channel
antagonist protects against neuronal loss after global
cerebral ischemia. Proc. Natl Acad. Sci. USA 90,
78947897 (1993).
32. Newman, D.J. & Cragg, G.M. Natural products as
sources of new drugs over the last 25years. J.Nat.
Prod. 70, 461477 (2007).
Describes the successes of natural products as a
source for new drugs.
33. Deacon, C.F. Therapeutic strategies based on
glucagon-like peptide-1. Diabetes 53, 21812189
(2004).
34. Von Itzstein, M. etal. Rational design of potent
sialidase-based inhibitors of influenza virus
replication. Nature 363, 418423 (1993).
35. Weibel, E.K., Hadvary, P., Hochuli, E., Kupfer, E. &
Lengsfeld, H. Lipstatin, an inhibitor of pancreatic
lipase produced by Streptomyces toxytricini.
1. Producing organism, fermentation, isolation and
biological activity. J.Antibiot. 40, 10811086
(1987).
36. Kluter, D.J. New thrombopoietic growth factors.
Blood 109, 46074616 (2007).
37. Remuzzi, G. etal. New therapeutics that antagonize
endothelin: promises and frustrations. Nature Rev.
Drug Discov. 1, 9861001 (2002).
38. Wood, J.M. etal. Structure-based design of aliskiren,
a novel orally effective renin inhibitor. Biochem.
Biophys. Res. Commun. 308, 698705 (2003).
39. Pommier, Y. etal. Integrase inhibitors to treat HIV/
AIDS. Nature Rev. Drug Discov. 4, 236248 (2005).
40. Espeseth, A.S. etal. HIV-1 integrase inhibitors that
compete with the target DNA substrate define a
unique strand transfer conformation for integrase.
Proc. Natl Acad. Sci. USA 97, 1124411249
(2000).
41. Lichtner, R. B. etal. Signaling-inactive epidermal
growth factor receptor/ligand complexes in intact
carcinoma cells by quinazoline kinase inhibitors.
Cancer Res. 61, 57905795 (2001).
42. Barker, A.J. etal. Studies leading to the identification
of ZD1830 (Iressa
TM
): an orally active, selective
epidermal growth factor receptor tyrosine kinase
inhibitor targeted to the treatment of cancer. Bioorg.
Med. Chem. Lett. 11, 19111914 (2001).
43. Leader, B., Baca, Q.J. & Golan, D.E. Protein
therapeutics: a summary and pharmacological
classification. Nature Rev. Drug Discov. 7, 2139
(2008).
44. Wilhelm, S. etal. Discovery and development of
sorafenib: a multikinase inhibitor for treating cancer.
Nature Rev. Drug Discov 5, 835844 (2006).
45. Goke, R. etal. Exendin-4 is a high potency agonist and
truncated exendin-(939)-amide an antagonist at the
glucagon-like peptide 1-(736)-amide receptor of
insulin-secreting -cells. J.Biol. Chem. 268,
1965019655 (1993).
46. Alvaro, G. & Di Fabio, R. Neurokinin 1 receptor
antagonists current prospects.
Curr. Opin. Drug Discov. Dev. 10, 613621 (2007).
ANALYSI S
nrd_3480_jul11.indd 517 22/06/2011 10:16
7PMPQYPWPENGCT
VCTIGVOGEJCPKUO
H
Docosunol
Lovoirucoum
Lubirosono
Nuoqlinido
Pomirolus
Punoluzino
Punumido
Sinocuocbins
Zonisumido
1VJGTU G
Aminolovulinic
ucid
Duomycin
Nolurubino
Voroorn
#GEVVTCPURQTVGT
CEVXKV[
F
Lzoimibo
#GEVTGEGRVQTCEVKXKV[ D
Moduluo
rosonso
Ariiruzolo
Acivuo
rosonso
Cinuculco
Llrombouq
Pumoloon
Aroiun
8osonun
Conivuun
Muruviroc
lnbibi
rosonso
Turqo nuclour
rocoors
lulvosrun
Milorisono
#GEVKQP
EJCPPGNCEVKXKV[
E
Acumrosuo
Momunino
Vuroniclino
Ziconoido
#GEVGP\[OGCEVKXKV[ C
Azuciidino
Cilosuzol
londuurinux
Miqlusu
Niuzoxunido
Niisinono
Orlisu
Vorinosu
Obor onzymo
inbibiors
Cusolunqin
Linozolid
Puloqruvir
Pouumulin
Zunumivir
Microbiul
onzymo inbibiors
Suroorin
Lnzymo
colucor
Proouso
inbibiors
Alisliron
Siuqliin
8orozomib
Kinuso
inbibiors
Goinib
lmuinib
Sorulonib
Suniinib
Sirolimus
(desloratadine)
61
, antimuscarinics (tiotropium)
62

and angiotensin receptor blockers (candesartan)
63,64
.
Conversely, in drug classes with mechanism-based tox-
icity, MMOAs that increase the therapeutic index have
been identified, as illustrated by SERMs such as ralox-
ifene
14,65
. A decrease in the number of iterations required
to identify an optimal MMOA for first-in-class drugs
could accelerate lead discovery and reduce late-stage
attrition, thereby increasing R&D productivity.
With regard to the discovery of follower drugs, the
opposite trend was seen compared to first-in-class
drugs, with target-based approaches accounting for 83
(51%) of these NMEs and phenotypic-based approaches
accounting for 30 (18%) NMEs. The reversal of the
trend is presumably the result of drug developers tak-
ing advantage of knowledge of a previously identified
MMOA to effectively use target-based tools. The tim-
ing of the use of these tools may also be important. A
recent report by DiMasi and Faden
66
on follower drugs
shows that research on a large percentage of follower
drugs was initiated before first-in-class approval. The
authors
66
concluded that drug development can often
be characterized as a race in which several firms pursue
investigational drugs with similar chemical structures
or with the same mechanism of action before any drug
in the class obtains regulatory marketing approval.
That is, it appears that once a mechanism of action or
a chemical class with the potential to be developed into
a drug is discovered, multiple organizations within the
pharmaceutical industry may pursue it vigorously. In
drug discovery, this race may contribute to the escalat-
ing costs, as there is only room for a few drugs in a class.
Additionally, the analysis by DiMasi and Faden
66
only
captures the drug classes that have been approved; if
the costs for organizations involved in a race around a
hypothesis that was later proven to be incorrect are also
considered, the total costs could be substantially higher.
The increased reliance on hypothesis-driven target-
based approaches in drug discovery has coincided with
the sequencing of the human genome and an apparent
Figure 4 | Activities of first-in-class small-molecule new molecular entities. Nearly half (22 out of 50) of the
first-in-class small-molecule drugs that were approved between 1999 and 2008 affected enzyme activity (a). The
molecular mechanisms of action (MMOAs) of these drugs included reversible, irreversible, competitive and
noncompetitive inhibition, blocking activation and stabilizing the substrate. The next largest group of targets (10 drugs)
were receptors (b), most of which were G protein-coupled receptors. Their MMOAs included agonism, partial agonism,
antagonism and allosteric modulation. Two drugs fulvestrant and mifepristone targeted nuclear receptors. Four of
the drugs targeted ion channel activity (c); their MMOAs included uncompetitive antagonism and partial agonism. One
drug, ezetimibe, targeted the activity of a transporter (d). The remaining drugs had other activities (e), or unclear targets
or MMOAs (f). Of the NMEs with other activities, two had a unique MMOA: verteporfin, a porphyrin that catalyses the
generation of reactive oxygen species and is used for photodynamic therapy; and daptomycin, which has an MMOA that
involves disruption of bacterial membranes. For details of the discovery and activities of each drug, see Supplementary
information S2 (box). *Sirolimus binds to the protein FKBP12 and the sirolimusFKBP12 complex inhibits the kinase
activity of mammalian target of rapamycin, whereas the other four kinase inhibitors target receptor tyrosine kinases.
Bortezomib inhibits the 26S proteasome a multiprotein complex by inhibiting the chymotryptic-like activity of the
proteasome.
Fulvestrant acts by promoting receptor degradation.

ANALYSI S
nrd_3480_jul11.indd 516 22/06/2011 10:16
belief by some that every target can provide the basis
for a drug. As such, research across the pharmaceutical
industry as well as academic institutions has increas-
ingly focused on targets, arguably at the expense of the
development of preclinical assays that translate more
effectively into clinical effects in patients with a specific
disease. In our analysis, we found that there are numerous
diverse MMOAs for approved new first-in-class drugs,
but drug discovery at present appears to be dominated by
a one size fits all approach, in which drugs are optimized
for binding affinity with less consideration for binding
kinetics and conformation. For optimal application of
target-based approaches, it is important to consider how
efficiently binding is coupled to the response (BOX2).
However, the molecular descriptors for the coupling fac-
tors may not be accurately captured by only consider-
ing binding affinity. Furthermore, an excessive focus on
affinity at a given target could lead to compromises being
made in pharmacokinetic properties that are critical for
the success of drugs, which has also been postulated to
be an underlying factor for current attritionrates
67
.
Reducing the impact of technical uncertainty on the
later, more costly stages of drug development through a
quick win/fast fail strategy has been proposed as a solu-
tion to the current problems with R&D productivity
2
.
However, this strategy does not address the key issues
that contribute to the greater technical uncertainty
and associated risk of failure. Our analysis leads us to
conclude that the identification of an optimal MMOA
has been a key factor contributing to the success of
phenotypic screening in the discovery of the first-in-
class NMEs in the 10-year period we studied. Thus, we
consider that technical risk and, consequently, overall
attrition in drug development could be decreased for
first-in-class drugs through the development and greater
use of translational phenotypic assays, and by considering
diverse MMOAs when using a target-based, hypothesis-
driven strategy.
1. Munos, B. Lessons for 60years of pharmaceutical
(2009).
2. Paul, S.M. etal. How to improve R.&D productivity:
the pharmaceutical industrys grand challenge. Nature
Rev. Drug Discov. 9, 203214 (2010).
3. Lindsay, M.A. Target discovery. Nature Rev. Drug
Discov. 2, 831838 (2003).
4. Imming, P., Sinning, C. & Meyer A. Drugs, their targets
and the nature and number of drug targets. Nature
Rev. Drug Discov. 5, 821834 (2006).
5. Overington, J.P., Al-Lazikani, B. & Hopkins, A.L. How
many drug targets are there? Nature Rev. Drug
Discov. 5, 993996(2006).
6. Williams, M. Systems and integrative biology as
alternative guises for pharmacology: prime time for an
iPharm concept? Biochem. Pharmacol. 70,
17071716 (2005).
7. Flordellis, C.S., Manolis, A.S., Paris, H. & Karabinis, A.
Rethinking target discovery in polygenic diseases.
Curr. Top. Med. Chem. 6, 17911798 (2006).
8. Urban, J.D. etal. Functional selectivity and classical
concepts of quantitative pharmacology J.Pharmacol.
Exp. Ther. 320, 113 (2007).
Formalizes the concept of functional selectivity,
whereby multiple unique ligands can bind to one
receptor to initiate different responses.
9. Kenakin, T. & Miller, L.J. Seven transmembrane
receptors as shapeshifting proteins: the impact of
allosteric modulation and functional selectivity on
drug discovery. Pharmacol. Rev. 62, 265304
(2010).
10. Swinney, D.C. Biochemical mechanisms of drug
action: what does it take for success? Nature Rev.
Drug Discov. 3, 801808 (2004).
Describes how the MMOA influences the
therapeutic index and utility of a medicine and
introduces biochemical efficiency as a metric to
quantify this influence.
11. Swinney, D.C. Biochemical mechanisms of new
molecular entities (NMEs) approved by United States
FDA during 20012004: mechanisms leading to
optimal efficacy and safety. Curr. Top. Med. Chem. 6,
461478 (2006).
12. Swinney, D.C. Applications of binding kinetics to drug
discovery: translation of binding mechanism to
clinically differentiated therapeutic responses. Pharm.
Med. 22, 2334 (2008).
13. Yun, C.H. etal. The T790M mutation in EGFR kinase
causes drug resistance by increasing the affinity for ATP.
Proc. Natl Acad. Sci. USA 105, 20702075 (2008).
Provides an illustration of how drug resistance
could be overcome through an understanding of
the MMOA.
14. Brzozowski, A.M. etal. Molecular basis of agonism
and antagonism in the oestrogen receptor. Nature
389, 753758 (1997).
Shows structurally how agonists and antagonists
bind at the same site but with different binding
modes that result in different responses.
15. Roth, G.J. & Majerus, P.W. The mechanism of the
effect of aspirin on human platelets. I. Acetylation of a
particulate fraction protein. J.Clin. Invest. 56,
624632 (1975).
16. Majerus, P.W., Broze, G.J. Jr, Miletich, J.P. & Tollefsen,
D.M. in Goodman & Gilmans The pharmacological
basis of therapeutics. (eds Hardman, J. G. & Limbird, L.
E.)1353 (McGraw-Hill, New York, 1996).
17. Copeland, R.A., Pompliano, D.L. & Meek, T.D.
Drug-target residence time and its implications for
lead optimization. Nature Rev. Drug Discov. 5,
730739 (2006).
18. Timmino, P.J. & Copeland, R.A. Residence time of
receptorligand complexes and its effect on biological
function. Biochemistry 47, 54815492 (2008).
19. Lu, H. & Tonge, P.J. Drug-target residence time:
critical information for lead optimization. Curr. Opin.
Chem. Biol. 14, 18 (2010).
20. Johnson, D.S., Weerapana, E. & Cravatt, B.F.
Strategies for discovering and derisking covalent,
irreversible enzyme inhibitors. Future Med. Chem. 2,
949964 (2010).
21. Ohlson, S. Designing transient binding drugs: a new
concept for drug discovery. Drug Discov. Today 13,
433439 (2008).
22. Lipton, S.A. Paradigm shift in neuroprotection by
NMDA receptor blockade: Memantine and beyond.
23. Lipton, S.A. Pathology activated therapeutics for
neuroprotection. Nature Rev. Neurosci. 8, 803808
(2007).
Describes the principle that drugs should be
activated by the pathological state that they are
intended to inhibit.
24. Changeux, J.P. Allosteric receptors: from electric
organ to cognition. Annu. Rev. Pharmacol. Toxicol. 50,
138 (2010).
25. Conn, J.P., Christopoulos, A. & Lindsley, C.W.
Allosteric modulators of GPCRs: a novel approach for
the treatment of CNS disorders. Nature Rev. Drug
Discov. 8, 4154 (2009).
26. Hanck, D.A. etal. Using lidocaine and benzocaine to
link sodium channel molecular conformations to state-
dependent antiarrhythic drug affinity. Circ. Res. 105,
492499 (2009).
27. Butterworth, J.F. & Strichartz, G.R. Molecular
mechanisms of local anesthesia: a review.
Anesthesiology 72, 711734 (1990).
28. Wilson, D.N. etal. The oxazolidinone antibiotics
perturb the ribosomal peptidyl-transferase center and
effect tRNA positioning. Proc. Natl Acad. Sci. USA
105, 1333913344 (2008).
29. Nemeth, E.F. Misconceptions about calcimimetics.
Ann. NYAcad. Sci. 1068, 471476 (2006).
Discusses lessons learned in the discovery of
cinacalcet, with emphasis on the importance of
using an understanding of physiology.
30. Salisbury, B.G. etal. Hypocholesterolemic activity of a
novel inhibitor of cholesterol absorption, SCH 48461.
Athlerosclerosis 115, 4563 (1995).
31. Valentino, D. etal. A selective N-type calcium channel
antagonist protects against neuronal loss after global
cerebral ischemia. Proc. Natl Acad. Sci. USA 90,
78947897 (1993).
32. Newman, D.J. & Cragg, G.M. Natural products as
sources of new drugs over the last 25years. J.Nat.
Prod. 70, 461477 (2007).
Describes the successes of natural products as a
source for new drugs.
33. Deacon, C.F. Therapeutic strategies based on
glucagon-like peptide-1. Diabetes 53, 21812189
(2004).
34. Von Itzstein, M. etal. Rational design of potent
sialidase-based inhibitors of influenza virus
replication. Nature 363, 418423 (1993).
35. Weibel, E.K., Hadvary, P., Hochuli, E., Kupfer, E. &
Lengsfeld, H. Lipstatin, an inhibitor of pancreatic
lipase produced by Streptomyces toxytricini.
1. Producing organism, fermentation, isolation and
biological activity. J.Antibiot. 40, 10811086
(1987).
36. Kluter, D.J. New thrombopoietic growth factors.
Blood 109, 46074616 (2007).
37. Remuzzi, G. etal. New therapeutics that antagonize
endothelin: promises and frustrations. Nature Rev.
Drug Discov. 1, 9861001 (2002).
38. Wood, J.M. etal. Structure-based design of aliskiren,
a novel orally effective renin inhibitor. Biochem.
Biophys. Res. Commun. 308, 698705 (2003).
39. Pommier, Y. etal. Integrase inhibitors to treat HIV/
AIDS. Nature Rev. Drug Discov. 4, 236248 (2005).
40. Espeseth, A.S. etal. HIV-1 integrase inhibitors that
compete with the target DNA substrate define a
unique strand transfer conformation for integrase.
Proc. Natl Acad. Sci. USA 97, 1124411249
(2000).
41. Lichtner, R. B. etal. Signaling-inactive epidermal
growth factor receptor/ligand complexes in intact
carcinoma cells by quinazoline kinase inhibitors.
Cancer Res. 61, 57905795 (2001).
42. Barker, A.J. etal. Studies leading to the identification
of ZD1830 (Iressa
TM
): an orally active, selective
epidermal growth factor receptor tyrosine kinase
inhibitor targeted to the treatment of cancer. Bioorg.
Med. Chem. Lett. 11, 19111914 (2001).
43. Leader, B., Baca, Q.J. & Golan, D.E. Protein
therapeutics: a summary and pharmacological
classification. Nature Rev. Drug Discov. 7, 2139
(2008).
44. Wilhelm, S. etal. Discovery and development of
sorafenib: a multikinase inhibitor for treating cancer.
Nature Rev. Drug Discov 5, 835844 (2006).
45. Goke, R. etal. Exendin-4 is a high potency agonist and
truncated exendin-(939)-amide an antagonist at the
glucagon-like peptide 1-(736)-amide receptor of
insulin-secreting -cells. J.Biol. Chem. 268,
1965019655 (1993).
46. Alvaro, G. & Di Fabio, R. Neurokinin 1 receptor
antagonists current prospects.
Curr. Opin. Drug Discov. Dev. 10, 613621 (2007).
ANALYSI S
nrd_3480_jul11.indd 517 22/06/2011 10:16
7PMPQYPWPENGCT
VCTIGVOGEJCPKUO
H
Docosunol
Lovoirucoum
Lubirosono
Nuoqlinido
Pomirolus
Punoluzino
Punumido
Sinocuocbins
Zonisumido
1VJGTU G
Aminolovulinic
ucid
Duomycin
Nolurubino
Voroorn
#GEVVTCPURQTVGT
CEVXKV[
F
Lzoimibo
#GEVTGEGRVQTCEVKXKV[ D
Moduluo
rosonso
Ariiruzolo
Acivuo
rosonso
Cinuculco
Llrombouq
Pumoloon
Aroiun
8osonun
Conivuun
Muruviroc
lnbibi
rosonso
Turqo nuclour
rocoors
lulvosrun
Milorisono
#GEVKQP
EJCPPGNCEVKXKV[
E
Acumrosuo
Momunino
Vuroniclino
Ziconoido
#GEVGP\[OGCEVKXKV[ C
Azuciidino
Cilosuzol
londuurinux
Miqlusu
Niuzoxunido
Niisinono
Orlisu
Vorinosu
Obor onzymo
inbibiors
Cusolunqin
Linozolid
Puloqruvir
Pouumulin
Zunumivir
Microbiul
onzymo inbibiors
Suroorin
Lnzymo
colucor
Proouso
inbibiors
Alisliron
Siuqliin
8orozomib
Kinuso
inbibiors
Goinib
lmuinib
Sorulonib
Suniinib
Sirolimus
(desloratadine)
61
, antimuscarinics (tiotropium)
62

and angiotensin receptor blockers (candesartan)
63,64
.
Conversely, in drug classes with mechanism-based tox-
icity, MMOAs that increase the therapeutic index have
been identified, as illustrated by SERMs such as ralox-
ifene
14,65
. A decrease in the number of iterations required
to identify an optimal MMOA for first-in-class drugs
could accelerate lead discovery and reduce late-stage
attrition, thereby increasing R&D productivity.
With regard to the discovery of follower drugs, the
opposite trend was seen compared to first-in-class
drugs, with target-based approaches accounting for 83
(51%) of these NMEs and phenotypic-based approaches
accounting for 30 (18%) NMEs. The reversal of the
trend is presumably the result of drug developers tak-
ing advantage of knowledge of a previously identified
MMOA to effectively use target-based tools. The tim-
ing of the use of these tools may also be important. A
recent report by DiMasi and Faden
66
on follower drugs
shows that research on a large percentage of follower
drugs was initiated before first-in-class approval. The
authors
66
concluded that drug development can often
be characterized as a race in which several firms pursue
investigational drugs with similar chemical structures
or with the same mechanism of action before any drug
in the class obtains regulatory marketing approval.
That is, it appears that once a mechanism of action or
a chemical class with the potential to be developed into
a drug is discovered, multiple organizations within the
pharmaceutical industry may pursue it vigorously. In
drug discovery, this race may contribute to the escalat-
ing costs, as there is only room for a few drugs in a class.
Additionally, the analysis by DiMasi and Faden
66
only
captures the drug classes that have been approved; if
the costs for organizations involved in a race around a
hypothesis that was later proven to be incorrect are also
considered, the total costs could be substantially higher.
The increased reliance on hypothesis-driven target-
based approaches in drug discovery has coincided with
the sequencing of the human genome and an apparent
Figure 4 | Activities of first-in-class small-molecule new molecular entities. Nearly half (22 out of 50) of the
first-in-class small-molecule drugs that were approved between 1999 and 2008 affected enzyme activity (a). The
molecular mechanisms of action (MMOAs) of these drugs included reversible, irreversible, competitive and
noncompetitive inhibition, blocking activation and stabilizing the substrate. The next largest group of targets (10 drugs)
were receptors (b), most of which were G protein-coupled receptors. Their MMOAs included agonism, partial agonism,
antagonism and allosteric modulation. Two drugs fulvestrant and mifepristone targeted nuclear receptors. Four of
the drugs targeted ion channel activity (c); their MMOAs included uncompetitive antagonism and partial agonism. One
drug, ezetimibe, targeted the activity of a transporter (d). The remaining drugs had other activities (e), or unclear targets
or MMOAs (f). Of the NMEs with other activities, two had a unique MMOA: verteporfin, a porphyrin that catalyses the
generation of reactive oxygen species and is used for photodynamic therapy; and daptomycin, which has an MMOA that
involves disruption of bacterial membranes. For details of the discovery and activities of each drug, see Supplementary
information S2 (box). *Sirolimus binds to the protein FKBP12 and the sirolimusFKBP12 complex inhibits the kinase
activity of mammalian target of rapamycin, whereas the other four kinase inhibitors target receptor tyrosine kinases.
Bortezomib inhibits the 26S proteasome a multiprotein complex by inhibiting the chymotryptic-like activity of the
proteasome.
Fulvestrant acts by promoting receptor degradation.

ANALYSI S
nrd_3480_jul11.indd 516 22/06/2011 10:16
119. Chaitman, B.R. etal. Anti-ischemic effects and
long-term survival during ranolazine monotherapy
in patients with chronic severe angina. J.Am. Coll.
Cardiol. 43, 13751382 (2004).
120. Chaitman, B.R. etal. Effects of ranolazine with
atenolol, amlodipine, or diltiazem on exercise
tolerance and angina frequency in patients with severe
chronic angina: a randomized controlled trial. JAMA
291, 309316 (2004).
121. Antzelevitch, C. etal. Electrophysiological effects of
ranolazine, a novel antianginal agent with antiarrhythmic
properties. Circulation 110, 904910 (2004).
122. Hunt, E. Pleuromutilin antibiotics. Drugs Future 25,
11631168 (2000).
123. Jain, K.K. An assessment of rufinamide as an anti-
epileptic in comparison with other drugs in clinical
development. Expert Opin. Investig. Drugs 9,
829840 (2000).
124. Rogawski, M.A. Diverse mechanisms of antiepileptic
drugs in the development pipeline. Epilepsy Res. 69,
273294 (2006).
125. Meltzer, S.M., Monk, B.J. & Tewari, K.S. Green tea
catechins for treatment of external genital warts. Am.
J.Obstet. Gynecol. 200, 233.e1233.e7 (2009).
126. Vezina, C., Kudelski, A. & Shegal, S.N. Rapamycin.
(AY-22,989), a new antifungal antibiotic. I. Taxomony
of the producing streptomycete and isolation of the
active principle. J.Antibiot. 10, 721726 (1975).
127. Richon, V.M. etal. Second generation hybrid polar
compounds are potent inducers of transformed cell
differentiation. Proc. Natl Acad. Sci. USA 93,
57055708 (1996).
128. Marks, P. A. & Breslow, R. Dimethyl sulfoxide to
vorinostat: development of this histone deacetylase
inhibitor as an anticancer drug.Nature Biotech.25,
8490 (2007).
129. Masuda Y. etal. 3-Sulfamoylmethyl-1,2-benzisoxazole,
a new type of anticonvulsant drug: pharmacological
profile. Arzneimittelforschung 30, 477483 (1980).
130. Maibaum, J. etal. Structural modification of the P2
position of 2,7-dialkyl-substituted
5(S)-amino-4(S)-hydroxy-8-phenyl-octanecarboxamides:
the discovery of aliskiren, a potent non-peptide human
renin inhibitor active after once daily dosing in
marmosets. J.Med. Chem. 50, 48324844 (2007).
131. Goldberg, A. in Cancer Drug Discovery and
Development: Proteasome Inhibitors in Cancer Therapy
(ed. Adams, J.) 1738 (Humana, Totowa, 2004).
132. Stein, R.L., Ma, Y.T. & Brand, S. Inhibitors of the 26s
proteolytic complex and the 20s proteasome
contained therein. US Patent 5,693,617 (1995).
133. Decaux, G., Soupart, A. & Vassart, G. Non-peptide
arginine-vasopressin antagonists: the vaptans. Lancet
371, 16241632 (2008).
134. Flexner, C. HIV drug development: the next 25years.
135. Tsibris, A.M. & Kuritzkes, D.R. Chemokine
antagonists as therapeutics: focus on HIV-1. Annu.
Rev. Med. 58, 445459 (2007).
136. Dorr, P. etal. Maraviroc (UK-427,857), a potent,
orally bioavailable, and selective small-molecule
inhibitor of chemokine receptor CCR5 with broad-
spectrum anti-human immunodeficiency virus type1
activity. Antimicrob. Agents Chemother. 49,
47214732 (2005).
137. Watson, C., Jenkinson, S., Kazmierski, W. & Kenakin, T.
The CCR5 receptor-based mechanism of action of
873140, a potent allosteric noncompetitive HIV entry
inhibitor. Mol. Pharmacol. 67, 12681282 (2005).
138. Pincus, G. (ed.) The Control of Fertility. 128138
(Academic Press, New York, 1965).
139. Belanger, A., Philibert, D. & Teutsch, G. Regio and
stereospecific synthesis of 11-substituted
19-norsteroids. Steroids 37, 361382 (1981).
140. Mahajan, D.K. & London, S.N. Mifepristone
(RU486): a review. Fertil. Steril. 68, 967976 (1997).
141. Raaijmakers, H.C., Versteegh, J.E. & Uitdehaag, J.C.
The X-ray structure of RU486 bound to the progesterone
receptor in a destabilized agonistic conformation. J.Biol.
Chem. 284, 1957219579 (2009).
142. Hadvary, P., Lengsfeld, H. & Wolfer, H. Inhibition of
pancreatic lipase invitro by the covalent inhibitor
tetrahydrolipstatin. Biochem. J. 256, 357361 (1988).
143. Hazuda, D.J. etal. Inhibitors of strand transfer that
prevent integration and inhibit HIV-1 replication in
cells. Science 287, 646650 (2000).
144. Summa, V. et al. Discovery of raltegravir. A potent,
selective orally bioavailable HIV-integrase inhibitor for
the treatment of HIV-AIDS infection. J.Med. Chem.
51, 58435855 (2008).
145. Buysse, D., Bate, G. & Kirkpatrick, P. Ramelteon.
146. Drucker, D.J. The biology of incretin hormones. Cell.
Metab. 3, 153165 (2006).
147. Cohen, H.T. & McGovern, F.J. Renal-cell carcinoma.
N.Engl. J.Med. 353, 24772490 (2005).
148. Atkins, M.B. etal. Innovations and challenges in renal
cancer: consensus statement from the first international
conference. Clin. Cancer Res. 9, 6277S6281S (2004).
149. Bergers, G. etal. Benefits of targeting both pericytes
and endothelial cells in the tumor vasculature with
kinase inhibitors. J.Clin. Invest. 111, 12871295
(2003).
150. Mendel, D.B. etal. Invivo anti-tumor activity of
SU11248, a novel tyrosine kinase inhibitor targeting
VEGF and PDGF receptors: determination of a
pharmacokinetic/pharmacodynamic relationship.
Clin. Cancer Res. 9, 327337 (2003).
151. De Clercq, E. Strategies in the design of antiviral
drugs. Nature Rev. Drug Discov. 1, 1325 (2002).
152. Boismare, F. etal. A homotaurine derivative reduces
the voluntary intake of ethanol by rats: are cerebral
GABA receptors involved? Pharmacol. Biochem.
Behav. 21, 787789 (1984).
153. Kennedy, J.C., Pottier, R.H. & Pross, D.C.
Photodynamic therapy with endogenous protoporphyrin
IX: basic principles and present clinical experience.
J.Photochem. Photobiol. B 6, 143148 (1990).
154. Sima, A.A. F., Kennedy, J.C., Blakeslee, D. &
Robertson, D.M. Experimental porphyric neuropathy:
a preliminary report. Can. J.Neurol. Sci. 8 105114
(1981).
155. Choay, J. etal. Structureactivity relationship in
heparin: a synthetic pentasaccharide with high affinity
for antithrombin III and eliciting high anti-factor Xa
activity. Biochem. Biophys. Res. Commun. 116,
492499 (1983).
156. Hirsh, J. etal. Heparin and low-molecular-weight
heparin: mechanisms of action, pharmacokinetics,
dosing, monitoring, efficacy, and safety. Chest 119,
64S94S (2001).
157. Walenga, J.M. etal. Development of a synthetic
heparin pentasaccharide: fondaparinux. Turk.
J.Haematol. 19, 137150 (2002).
158. Blau, N. & Erlandsen, H. The metabolic and molecular
bases of tetrahydrobiopterin-responsive phenylalanine
hydroxylase deficiency. Mol. Genet. Metab. 82,
101111 (2004).
159. Niederwieser, A. & Curtius, H.C. in Inherited Diseases
of Amino Acid Metabolism (eds Bickel, H. & Wachtel,
U.) 104121 (Georg Thieme, Stuttgart, 1985).
160. Kure, S. etal. Tetrahydrobiopterin-responsive
phenylalanine hydroxylase deficiency. J.Pediatr. 135,
375378 (1999).
161. Muntau, A.C. etal. Tetrahydrobiopterin as an
alternative treatment for mild phenylketonuria.
N.Engl. J.Med. 347, 21222132 (2002).
162. Mellish, K.J. & Brown, S.B. Verteporfin: a milestone
in ophthalmology and photodynamic therapy.
Expert Opin. Pharmacother. 2, 351361
(2001).
Acknowledgements
We wish to acknowledge the employees of Roche (Palo Alto)
who created a great environment to do drug discovery
research. We specifically thank the members of the
Biochemical Pharmacology Core led by A. Ford and the
Virology Disease Biology Area led by N. Cammack for their
support and encouragement. D.C.S. also wishes to thank the
many scientists whose feedback, constructive criticism and
desire to discover new medicines helped to provide the moti-
vation for this work.
Competing interests statement:
The authors declare competing financial interests: see Web
FURTHER INFORMATION
Drugs@FDA: http://www.accessdata.fda.gov/scripts/cder/
drugsatfda
See online article: S1 (table) | S2 (box)
ALL LINKS ARE ACTIVE IN THE ONLINE PDF
ANALYSI S
nrd_3480_jul11.indd 519 22/06/2011 10:16
47. Wijayaratne, A.L. & McDonnell, D.P. The human
estrogen receptor- is a ubiquitinated protein whose
stability is affected differentially by agonists, antagonists,
and selective estrogen receptor modulators. J.Biol.
Chem. 276, 3568435692 (2001).
48. Ferrara, N. etal. Discovery and development of
bevacizumab, an anti-VEGF antibody for treating
cancer. Nature Rev. Drug Discov. 3, 391400
(2004).
49. Capdeville, R., Buchdunger, E., Zimmermann, J. &
Matter A. Glivec (ST571, imatinib), a rationally
developed, targeted anticancer drug. Nature Rev.
Drug. Discov. 1, 493502 (2002).
50. Wacker, D. etal. Conserved binding mode of human
2 adrenergic receptor inverse agonists and
antagonist revealed by X-ray crystallography. J.Am.
Chem. Soc. 132, 1144311445 (2010).
51. Lemmon, M.A. & Schlessinger, J. Cell signaling by
receptor tyrosine kinases. Cell 141, 11171134
(2010).
52. Li, P. etal. Cytochromec and dATP-dependent
formation of Apaf-1/caspase-9 complex initiates an
apoptotic protease cascade. Cell 91, 479489.
53. Johnson, K.A. Role of induced fit in enzyme
specificity: a molecular forward/reverse switch. J.Biol.
Chem. 283, 2629726301 (2008).
54. Sigoillot, F.D. & King, R.W. Vigilance and validation:
keys to success in RNAi screening. ACS Chem. Biol. 6,
4760 (2011).
55. Hergenrother, P.J. & Palchaudhuri, R. Transcript
profiling and RNA interference as tools to identify
small molecule mechanisms and therapeutic potential.
ACS Chem. Biol. 6, 2133 (2011).
56. Macarron, R. et al. Impact of high-throughput
screening in biomedical research. Nature Rev. Drug.
Discov. 10, 188195 (2011).
57. Pruss, R.M. Phenotypic screening strategies for
neurodegenerative diseases: a pathway to discover
novel drug candidates and potential disease targets or
mechanisms. CNS Neurol. Disord. Drug Targets 9,
693700 (2010).
58. Bickle, M. The beautiful cell: high-content screening in
drug discovery. Anal. Bioanal. Chem. 398, 219226
(2010).
59. Mayer, A.M. et al. The odyssey of marine
pharmaceuticals: a current pipeline perspective.
Trends Pharmacol. Sci. 31, 255265 (2010).
60. Telling, J.L. et al. The biological activity of human
CD20 monoclonal antibodies is linked to unique
epitopes on CD20. J.Immunol. 177, 362371
(2006).
61. Anthes, J.C. etal. Biochemical characterization of
desloratadine, a potent antagonist of the human
histamine H
1
receptor. Eur. J.Pharmacol. 449,
229237 (2002).
62. Disse, B. etal. Tiotropium (Spiriva): mechanistic
considerations and clinical profile in obstructive lung
disease. Life Sci. 64, 457464 (1999).
63. Vauquelin, G., Fierens, F. & Van Liefde, I. Long-lasting
AT1 receptor binding and protection by candesartan:
comparison to other biphenyl-tetrazole sartans.
J. Hypertens. 24, S23S30 (2006).
64. Fuchs, B. etal. Comparative pharmacodynamics and
pharmacokinetics of candesartan and losartan in man.
J.Pharm. Pharmacol. 52, 10751083 (2000).
65. Gustafsson, J. A. Raloxifene: magic bullet for heart
and bone? Nature Med. 4, 152153 (1998).
66. DiMasi, J.A. & Fadon, L.B. Competitiveness in follow-
on drug R&D: a race or imitation? Nature Rev. Drug
Discov. 10, 15 (2011).
67. Gleeson, M.P., Hersey, A., Montanari, D. &
Overington, J. Probing the links between invitro
potency, ADMET and physiocochemical parameters.
68. Fersht, A. Enzyme Structure and Mechanism 88109
(W. H Freeman and Company, New York,1985).
69. Issa, J.P. J., Kantarjian, H. M. & Kirkpatrick, P.
Azacitidine. Nature Rev. Drug Discov. 4, 275276
(2005).
70. Martel, R.R., Klicius, J. & Galet, S. Inhibition of
immune response by rapamycin, a new antifungal
antibiotic. Can. J.Physiol. Pharmacol. 55, 4851
(1977).
71. Bartizal, K. etal. Invitro antifungal activities and
invivo efficacies of 1,3--glucan synthesis inhibitors
L671,329, L646,991, tetrahydroechinocandin B, and
L687,781, a papulacandin. Antimicrob. Agents
Chemother. 36, 16481657 (1992).
72. Uchikawa, O. etal. Synthesis of a novel series of
tricyclic indan derivatives as melatonin receptor
agonists. J.Med. Chem. 45, 42224239 (2002).
73. Kenakin, T. Pharmacologic Analysis of DrugReceptor
Interaction 242395 (Lippincott-Raven Publishers,
Philadelphia, 1997).
74. Burris K.D. etal. Aripiprazole, a novel antipsychotic,
is a high-affinity partial agonist at human dopamine
D2 receptors. J.Pharmacol. Exp. Ther. 302,
381389 (2002).
75. Pulvirenti, L. & Koob, G.F. Dopamine receptor
agonists, partial agonists and psychostimulant
addiction. Trends Pharmacol. Sci. 15, 374379
(1994).
76. Coe, J.E. etal. Varenicline: an 42 nicotinic receptor
partial agonist for smoking cessation. J.Med. Chem.
48, 34743477 (2005).
Describes the thinking that led to a
mechanism-based search for a partial agonist of
nicotinic receptors.
77. Rickter, A.M. etal. Preliminary studies on a more
effective phototoxic agent than hematoporphyrin.
J.Natl Cancer Inst. 79, 13271332 (1987).
78. Hemphill, A., Mueller, J. & Esposito, M. Nitazoxanide,
a broad-spectrum thiazolide anti-infective agent for
the treatment of gastrointestinal infections. Expert
Opin. Pharmacother. 7, 953964 (2006).
79. Rossignol, J.F. & Maisonneuve, H. Nitazoxanide in the
treatment of Teania saginata and Hymenolepis nana
infections. Am. J.Trop. Med. Hyg. 33, 511512
(1984).
80. Lewis, D.A. & Lieberman, J.A. Catching up on
schizophrenia: natural history and neurobiology.
Neuron 28, 325334 (2000).
81. Yasuda, Y. etal. 7-[3-[4-(2,3 dimethylphenyl)
piperazinyl] propoxy]-2(1H)-quinolinone (OPC-4392),
a presynaptic dopamine receptor agonist and
postsynaptic D2 receptor antagonist. Life Sci. 42,
19411954 (1988).
82. Kikuchi, T. etal. 7-(4-[4-(2,3-dichlorophenyl)
-1-piperazinyl]butyloxy)-3,4-dihydro-2(1H)-
quinolinone (OPC-14597), a new putative
antipsychotic drug with both presynaptic dopamine
autoreceptor agonistic activity and postsynaptic D2
receptor antagonistic activity. J.Pharmacol. Exp. Ther.
274, 329336 (1995).
83. Oshiro, Y. etal. Novel antipsychotic agents with
dopamine autoreceptor agonist properties: synthesis
and pharmacology of 7-[4-(4-phenyl-1-piperazinyl)
butoxy]-3,4-dihydro-2(1H)-quinolinone derivatives.
J.Med. Chem. 41, 658667 (1998).
84. Inoue, T., Domae, M., Yamada, K. & Furukawa, T.
Effects of the novel antipsychotic agent
7-(4-[4-(2,3-dichorophenyl)-1-piperazinyl]
butyloxy)-3,4-dihydro-2(1H)-quinoline (OPC-14597)
on prolactin release from the rat anterior pituitary
gland. J.Pharmacol. Exp. Ther. 277, 137143
(1996).
85. Egger, G., Liang, G., Aparicio, A. & Jones, P.A.
Epigenetics in human disease and prospects for
epigenetic therapy. Nature 429, 457463
(2004).
86. Satistowska-Schroder, E.T., Kerridge, D. & Perry, H.
Echinocandin inhibition of 1,3--D-glucan synthase
from Candida albicans. FEBS Lett. 173, 134138
(1984).
87. Nishi, T. etal. Studies on 2-oxoquinoline derivatives as
blood platelet aggregation inhibitors. I. Alkyl
4-(2-oxo-1,2,3,4-tetrahydro-6-quinolyloxy)butyrates
and related compounds. Chem. Pharm. Bull. 31,
798810 (1983).
88. Tally, F.P. & DeBruin M.F. Development of daptomycin
for Gram-positive infections. J.Antimicrob. Chemother.
46, 523526 (2000).
89. Stock, C, C. & Francis, T.J. The inactivation of the
virus of epidemic influenza by soaps. J.Exp. Med. 71,
661681 (1940).
90. Snipes, W., Person, S., Keller, G., Taylor, W. & Keith, A.
Inactivation of lipid-containing viruses by long-chain
alcohols. Antimicrob. Agents Chemother. 11, 98104
(1977).
91. Sands, J., Auperin, D. & Snipes, W. Extreme sensitivity
of enveloped viruses including herpes-simplex, to
long-chain unsaturated monglycerides and alcohols.
Antimicrob. Agents Chemother. 15, 6773
(1979).
92. Katz, D.H., Marcelletti, J.F., Khalil, M.H., Pope, L.E.
& Katz, L.E. Antiviral activity of 1-docosanol, an
inhibitor of lipid-enveloped viruses including herpes
simplex. Proc. Natl Acad. Sci. USA 88, 1082510829
(1991).
93. Wakeling, A.E., Dukes, M. & Bowler, J. A potent
specific pure antiestrogen with clinical potential.
Cancer Res. 51, 38673873 (1991).
94. Stenoien, D.L. etal. FRAP reveals that mobility of
oestrogen receptor- is ligand- and proteasome-
dependent. Nature Cell Biol. 3, 1523 (2001).
95. Glower, A.J., Noyer, M., Verloes, R., Gobert, J. &
Wulfert, E. UCB L059, a novel anti-convulsant drug:
pharmacological profile in animals. Eur. J.Pharmacol.
222, 193203 (1992).
96. Shinabarger, D. Mechanism of action of the
oxazolidinone antibacterial agents. Expert Opin.
Investig. Drugs 8, 11951202 (1999).
97. Brickner, S.J. Oxazolidinone antibacterial agents.
Curr. Pharm. Des. 2, 175194 (1996).
98. Cuppoletti J. etal. Recombinant and native intestinal
cell ClC-2 Cl
channels are activated by RU-0211.

Gastroenterology 122, A538 (2002).
99. Cuppoletti, J. etal. SPI-0211 activates T84 cell
chloride transport and recombinant human ClC-2
chloride currents. Am. J.Physiol. 287, C1173C1183
(2004).
100. Pea-Mnzenmayer, G. etal. Basolateral localization
of native ClC-2 chloride channels in absorptive
intestinal epithelial cells and basolateral sorting
encoded by a CBS-2 domain di-leucine motif. J.Cell
Sci. 118, 42434252 (2005).
101. Parsons, C.G., Danysz, W. & Quack, G. Memantine is a
clinically well-tolerated N-methyl-d-aspartate (NMDA)
receptor antagonist a review of the preclinical data.
Neuropharmacology 38, 735767 (1999).
102. Gerzon, K. etal. The adamantyl group in medicinal
agents. I. Hypoglycemic N-arylsulfonyl-N-
adamantylureas. J.Med. Chem. 6, 760763 (1963).
103. Bormann, J. Memantine is a potent blocker of
N-methyl-d-aspartate (NMDA) receptor channels. Eur.
J.Pharmacol. 166, 591592 (1989).
104. Platt, F.M., Neises, G.R., Dwek, R.A. & Butters, T.D.
N-butyldeoxynojirimycin is a novel inhibitor of
glycolipid biosynthesis. J.Biol. Chem. 269,
83628365 (1994).
105. Pastores, G.M. & Barnett, N.L. Substrate reduction
therapy: miglustat as a remedy for symptomatic
patients with Gaucher disease type1. Expert Opin.
Investig. Drugs. 12, 273281 (2003).
106. Hu, S. etal. Pancreatic -cell KATP channel activity
and membrane-binding studies with nateglinide: a
comarison with sulfonylureas and repaglinide.
J.Pharmacol. Ther. 293, 444452 (2000).
107. Shinkai, H. etal. N-acylphenylanalines and related
compounds. A new class of oral hypoglycemic agents.
J.Med. Chem. 31, 20922097 (1988).
J.Med. Chem. 32, 14361441 (1989).
109. Parker, W.B. etal. Purine nucleoside analogues in
development for the treatment of cancer. Curr. Opin.
Investig. Drugs 5, 592596 (2004).
110. Rodriguez, C.O. etal. Mechanisms for T-cell selective
cytotoxicity of arabinosylguanine. Blood 102,
18421848 (2003).
111. Krenitsky, T.A. etal. An enzymatic synthesis of
purine-d-arabinonucleosides. Carbohydr. Res. 97,
139146 (1981).
112. Lambe, C.U. etal. 2-amino-6-methoxypurine
arabinoside: an agent for T-cell malignancies. Cancer
Res. 55, 33523356 (1995).
113. Gandhi, V., Keating, M.J., Bate, G. & Kirkpatrick, P.
Nelarabine. Nature Rev. Drug Discov. 5, 1718 (2006).
114. Lock, E.A. etal. From toxicological problem to
therapeutic use: the discovery of the mode of action
of 2-(2-nitro-4-trifluoromethylbenzoyl)-1,
3-cyclohexanedione (NTBC), its toxicology and
development as a drug. J.Inherit. Metab. Dis. 21,
498506 (1998).
115. Kavana, M. & Moran, G.R. Interaction of
(4-hydroxyphenyl)pyruvate dioxygenase with the
specific inhibitor 2-[2-Nitro-4-(trifluoromethyl)
benzoly]-1,3-cyclohexanedione. Biochemistry 42,
1023810245 (2003).
116. Brownlee, J.M., Johnson-Winters, K., Harrison, D.H. T.
& Moran, G.R. Structure of the ferrous form of
4-(hydroxyphenyl)pyruvate dehydrogenase from
Streptomyces avermitilis in complex with the
therapeutic herbicide, NTBC. Biochemistry 43,
63706377 (2004).
117. Yanagihara, Y., Kasai, H., Kawashima, T. & Shida, T.
Immunopharmacological studies on TBX, a new
antiallergic drug (1). Inhibitory effects on passive
cutaneous anaphylaxis in rats and guinea pigs.
Jpn. J.Pharmacol. 48, 91101 (1988).
118. Gaffney, S.M. Ranolazine, a novel agent for chronic
stable angina. Pharmacotherapy 26, 135142
(2006).
ANALYSI S
nrd_3480_jul11.indd 518 22/06/2011 10:16
119. Chaitman, B.R. etal. Anti-ischemic effects and
long-term survival during ranolazine monotherapy
in patients with chronic severe angina. J.Am. Coll.
Cardiol. 43, 13751382 (2004).
120. Chaitman, B.R. etal. Effects of ranolazine with
atenolol, amlodipine, or diltiazem on exercise
tolerance and angina frequency in patients with severe
chronic angina: a randomized controlled trial. JAMA
291, 309316 (2004).
121. Antzelevitch, C. etal. Electrophysiological effects of
ranolazine, a novel antianginal agent with antiarrhythmic
properties. Circulation 110, 904910 (2004).
122. Hunt, E. Pleuromutilin antibiotics. Drugs Future 25,
11631168 (2000).
123. Jain, K.K. An assessment of rufinamide as an anti-
epileptic in comparison with other drugs in clinical
development. Expert Opin. Investig. Drugs 9,
829840 (2000).
124. Rogawski, M.A. Diverse mechanisms of antiepileptic
drugs in the development pipeline. Epilepsy Res. 69,
273294 (2006).
125. Meltzer, S.M., Monk, B.J. & Tewari, K.S. Green tea
catechins for treatment of external genital warts. Am.
J.Obstet. Gynecol. 200, 233.e1233.e7 (2009).
126. Vezina, C., Kudelski, A. & Shegal, S.N. Rapamycin.
(AY-22,989), a new antifungal antibiotic. I. Taxomony
of the producing streptomycete and isolation of the
active principle. J.Antibiot. 10, 721726 (1975).
127. Richon, V.M. etal. Second generation hybrid polar
compounds are potent inducers of transformed cell
differentiation. Proc. Natl Acad. Sci. USA 93,
57055708 (1996).
128. Marks, P. A. & Breslow, R. Dimethyl sulfoxide to
vorinostat: development of this histone deacetylase
inhibitor as an anticancer drug.Nature Biotech.25,
8490 (2007).
129. Masuda Y. etal. 3-Sulfamoylmethyl-1,2-benzisoxazole,
a new type of anticonvulsant drug: pharmacological
profile. Arzneimittelforschung 30, 477483 (1980).
130. Maibaum, J. etal. Structural modification of the P2
position of 2,7-dialkyl-substituted
5(S)-amino-4(S)-hydroxy-8-phenyl-octanecarboxamides:
the discovery of aliskiren, a potent non-peptide human
renin inhibitor active after once daily dosing in
marmosets. J.Med. Chem. 50, 48324844 (2007).
131. Goldberg, A. in Cancer Drug Discovery and
Development: Proteasome Inhibitors in Cancer Therapy
(ed. Adams, J.) 1738 (Humana, Totowa, 2004).
132. Stein, R.L., Ma, Y.T. & Brand, S. Inhibitors of the 26s
proteolytic complex and the 20s proteasome
contained therein. US Patent 5,693,617 (1995).
133. Decaux, G., Soupart, A. & Vassart, G. Non-peptide
arginine-vasopressin antagonists: the vaptans. Lancet
371, 16241632 (2008).
134. Flexner, C. HIV drug development: the next 25years.
135. Tsibris, A.M. & Kuritzkes, D.R. Chemokine
antagonists as therapeutics: focus on HIV-1. Annu.
Rev. Med. 58, 445459 (2007).
136. Dorr, P. etal. Maraviroc (UK-427,857), a potent,
orally bioavailable, and selective small-molecule
inhibitor of chemokine receptor CCR5 with broad-
spectrum anti-human immunodeficiency virus type1
activity. Antimicrob. Agents Chemother. 49,
47214732 (2005).
137. Watson, C., Jenkinson, S., Kazmierski, W. & Kenakin, T.
The CCR5 receptor-based mechanism of action of
873140, a potent allosteric noncompetitive HIV entry
inhibitor. Mol. Pharmacol. 67, 12681282 (2005).
138. Pincus, G. (ed.) The Control of Fertility. 128138
(Academic Press, New York, 1965).
139. Belanger, A., Philibert, D. & Teutsch, G. Regio and
stereospecific synthesis of 11-substituted
19-norsteroids. Steroids 37, 361382 (1981).
140. Mahajan, D.K. & London, S.N. Mifepristone
(RU486): a review. Fertil. Steril. 68, 967976 (1997).
141. Raaijmakers, H.C., Versteegh, J.E. & Uitdehaag, J.C.
The X-ray structure of RU486 bound to the progesterone
receptor in a destabilized agonistic conformation. J.Biol.
Chem. 284, 1957219579 (2009).
142. Hadvary, P., Lengsfeld, H. & Wolfer, H. Inhibition of
pancreatic lipase invitro by the covalent inhibitor
tetrahydrolipstatin. Biochem. J. 256, 357361 (1988).
143. Hazuda, D.J. etal. Inhibitors of strand transfer that
prevent integration and inhibit HIV-1 replication in
cells. Science 287, 646650 (2000).
144. Summa, V. et al. Discovery of raltegravir. A potent,
selective orally bioavailable HIV-integrase inhibitor for
the treatment of HIV-AIDS infection. J.Med. Chem.
51, 58435855 (2008).
145. Buysse, D., Bate, G. & Kirkpatrick, P. Ramelteon.
146. Drucker, D.J. The biology of incretin hormones. Cell.
Metab. 3, 153165 (2006).
147. Cohen, H.T. & McGovern, F.J. Renal-cell carcinoma.
N.Engl. J.Med. 353, 24772490 (2005).
148. Atkins, M.B. etal. Innovations and challenges in renal
cancer: consensus statement from the first international
conference. Clin. Cancer Res. 9, 6277S6281S (2004).
149. Bergers, G. etal. Benefits of targeting both pericytes
and endothelial cells in the tumor vasculature with
kinase inhibitors. J.Clin. Invest. 111, 12871295
(2003).
150. Mendel, D.B. etal. Invivo anti-tumor activity of
SU11248, a novel tyrosine kinase inhibitor targeting
VEGF and PDGF receptors: determination of a
pharmacokinetic/pharmacodynamic relationship.
Clin. Cancer Res. 9, 327337 (2003).
151. De Clercq, E. Strategies in the design of antiviral
drugs. Nature Rev. Drug Discov. 1, 1325 (2002).
152. Boismare, F. etal. A homotaurine derivative reduces
the voluntary intake of ethanol by rats: are cerebral
GABA receptors involved? Pharmacol. Biochem.
Behav. 21, 787789 (1984).
153. Kennedy, J.C., Pottier, R.H. & Pross, D.C.
Photodynamic therapy with endogenous protoporphyrin
IX: basic principles and present clinical experience.
J.Photochem. Photobiol. B 6, 143148 (1990).
154. Sima, A.A. F., Kennedy, J.C., Blakeslee, D. &
Robertson, D.M. Experimental porphyric neuropathy:
a preliminary report. Can. J.Neurol. Sci. 8 105114
(1981).
155. Choay, J. etal. Structureactivity relationship in
heparin: a synthetic pentasaccharide with high affinity
for antithrombin III and eliciting high anti-factor Xa
activity. Biochem. Biophys. Res. Commun. 116,
492499 (1983).
156. Hirsh, J. etal. Heparin and low-molecular-weight
heparin: mechanisms of action, pharmacokinetics,
dosing, monitoring, efficacy, and safety. Chest 119,
64S94S (2001).
157. Walenga, J.M. etal. Development of a synthetic
heparin pentasaccharide: fondaparinux. Turk.
J.Haematol. 19, 137150 (2002).
158. Blau, N. & Erlandsen, H. The metabolic and molecular
bases of tetrahydrobiopterin-responsive phenylalanine
hydroxylase deficiency. Mol. Genet. Metab. 82,
101111 (2004).
159. Niederwieser, A. & Curtius, H.C. in Inherited Diseases
of Amino Acid Metabolism (eds Bickel, H. & Wachtel,
U.) 104121 (Georg Thieme, Stuttgart, 1985).
160. Kure, S. etal. Tetrahydrobiopterin-responsive
phenylalanine hydroxylase deficiency. J.Pediatr. 135,
375378 (1999).
161. Muntau, A.C. etal. Tetrahydrobiopterin as an
alternative treatment for mild phenylketonuria.
N.Engl. J.Med. 347, 21222132 (2002).
162. Mellish, K.J. & Brown, S.B. Verteporfin: a milestone
in ophthalmology and photodynamic therapy.
Expert Opin. Pharmacother. 2, 351361
(2001).
Acknowledgements
We wish to acknowledge the employees of Roche (Palo Alto)
who created a great environment to do drug discovery
research. We specifically thank the members of the
Biochemical Pharmacology Core led by A. Ford and the
Virology Disease Biology Area led by N. Cammack for their
support and encouragement. D.C.S. also wishes to thank the
many scientists whose feedback, constructive criticism and
desire to discover new medicines helped to provide the moti-
vation for this work.
Competing interests statement:
The authors declare competing financial interests: see Web
FURTHER INFORMATION
Drugs@FDA: http://www.accessdata.fda.gov/scripts/cder/
drugsatfda
See online article: S1 (table) | S2 (box)
ALL LINKS ARE ACTIVE IN THE ONLINE PDF
ANALYSI S
nrd_3480_jul11.indd 519 22/06/2011 10:16
47. Wijayaratne, A.L. & McDonnell, D.P. The human
estrogen receptor- is a ubiquitinated protein whose
stability is affected differentially by agonists, antagonists,
and selective estrogen receptor modulators. J.Biol.
Chem. 276, 3568435692 (2001).
48. Ferrara, N. etal. Discovery and development of
bevacizumab, an anti-VEGF antibody for treating
cancer. Nature Rev. Drug Discov. 3, 391400
(2004).
49. Capdeville, R., Buchdunger, E., Zimmermann, J. &
Matter A. Glivec (ST571, imatinib), a rationally
developed, targeted anticancer drug. Nature Rev.
Drug. Discov. 1, 493502 (2002).
50. Wacker, D. etal. Conserved binding mode of human
2 adrenergic receptor inverse agonists and
antagonist revealed by X-ray crystallography. J.Am.
Chem. Soc. 132, 1144311445 (2010).
51. Lemmon, M.A. & Schlessinger, J. Cell signaling by
receptor tyrosine kinases. Cell 141, 11171134
(2010).
52. Li, P. etal. Cytochromec and dATP-dependent
formation of Apaf-1/caspase-9 complex initiates an
apoptotic protease cascade. Cell 91, 479489.
53. Johnson, K.A. Role of induced fit in enzyme
specificity: a molecular forward/reverse switch. J.Biol.
Chem. 283, 2629726301 (2008).
54. Sigoillot, F.D. & King, R.W. Vigilance and validation:
keys to success in RNAi screening. ACS Chem. Biol. 6,
4760 (2011).
55. Hergenrother, P.J. & Palchaudhuri, R. Transcript
profiling and RNA interference as tools to identify
small molecule mechanisms and therapeutic potential.
ACS Chem. Biol. 6, 2133 (2011).
56. Macarron, R. et al. Impact of high-throughput
screening in biomedical research. Nature Rev. Drug.
Discov. 10, 188195 (2011).
57. Pruss, R.M. Phenotypic screening strategies for
neurodegenerative diseases: a pathway to discover
novel drug candidates and potential disease targets or
mechanisms. CNS Neurol. Disord. Drug Targets 9,
693700 (2010).
58. Bickle, M. The beautiful cell: high-content screening in
drug discovery. Anal. Bioanal. Chem. 398, 219226
(2010).
59. Mayer, A.M. et al. The odyssey of marine
pharmaceuticals: a current pipeline perspective.
Trends Pharmacol. Sci. 31, 255265 (2010).
60. Telling, J.L. et al. The biological activity of human
epitopes on CD20. J.Immunol. 177, 362371
(2006).
61. Anthes, J.C. etal. Biochemical characterization of
desloratadine, a potent antagonist of the human
histamine H
1
receptor. Eur. J.Pharmacol. 449,
229237 (2002).
62. Disse, B. etal. Tiotropium (Spiriva): mechanistic
considerations and clinical profile in obstructive lung
disease. Life Sci. 64, 457464 (1999).
63. Vauquelin, G., Fierens, F. & Van Liefde, I. Long-lasting
AT1 receptor binding and protection by candesartan:
comparison to other biphenyl-tetrazole sartans.
J. Hypertens. 24, S23S30 (2006).
64. Fuchs, B. etal. Comparative pharmacodynamics and
pharmacokinetics of candesartan and losartan in man.
J.Pharm. Pharmacol. 52, 10751083 (2000).
65. Gustafsson, J. A. Raloxifene: magic bullet for heart
and bone? Nature Med. 4, 152153 (1998).
66. DiMasi, J.A. & Fadon, L.B. Competitiveness in follow-
on drug R&D: a race or imitation? Nature Rev. Drug
Discov. 10, 15 (2011).
67. Gleeson, M.P., Hersey, A., Montanari, D. &
Overington, J. Probing the links between invitro
potency, ADMET and physiocochemical parameters.
68. Fersht, A. Enzyme Structure and Mechanism 88109
(W. H Freeman and Company, New York,1985).
69. Issa, J.P. J., Kantarjian, H. M. & Kirkpatrick, P.
Azacitidine. Nature Rev. Drug Discov. 4, 275276
(2005).
70. Martel, R.R., Klicius, J. & Galet, S. Inhibition of
immune response by rapamycin, a new antifungal
antibiotic. Can. J.Physiol. Pharmacol. 55, 4851
(1977).
71. Bartizal, K. etal. Invitro antifungal activities and
invivo efficacies of 1,3--glucan synthesis inhibitors
L671,329, L646,991, tetrahydroechinocandin B, and
L687,781, a papulacandin. Antimicrob. Agents
Chemother. 36, 16481657 (1992).
72. Uchikawa, O. etal. Synthesis of a novel series of
tricyclic indan derivatives as melatonin receptor
agonists. J.Med. Chem. 45, 42224239 (2002).
73. Kenakin, T. Pharmacologic Analysis of DrugReceptor
Interaction 242395 (Lippincott-Raven Publishers,
Philadelphia, 1997).
74. Burris K.D. etal. Aripiprazole, a novel antipsychotic,
is a high-affinity partial agonist at human dopamine
D2 receptors. J.Pharmacol. Exp. Ther. 302,
381389 (2002).
75. Pulvirenti, L. & Koob, G.F. Dopamine receptor
agonists, partial agonists and psychostimulant
addiction. Trends Pharmacol. Sci. 15, 374379
(1994).
76. Coe, J.E. etal. Varenicline: an 42 nicotinic receptor
partial agonist for smoking cessation. J.Med. Chem.
48, 34743477 (2005).
Describes the thinking that led to a
mechanism-based search for a partial agonist of
nicotinic receptors.
77. Rickter, A.M. etal. Preliminary studies on a more
effective phototoxic agent than hematoporphyrin.
J.Natl Cancer Inst. 79, 13271332 (1987).
78. Hemphill, A., Mueller, J. & Esposito, M. Nitazoxanide,
a broad-spectrum thiazolide anti-infective agent for
the treatment of gastrointestinal infections. Expert
Opin. Pharmacother. 7, 953964 (2006).
79. Rossignol, J.F. & Maisonneuve, H. Nitazoxanide in the
treatment of Teania saginata and Hymenolepis nana
infections. Am. J.Trop. Med. Hyg. 33, 511512
(1984).
80. Lewis, D.A. & Lieberman, J.A. Catching up on
schizophrenia: natural history and neurobiology.
Neuron 28, 325334 (2000).
81. Yasuda, Y. etal. 7-[3-[4-(2,3 dimethylphenyl)
piperazinyl] propoxy]-2(1H)-quinolinone (OPC-4392),
a presynaptic dopamine receptor agonist and
postsynaptic D2 receptor antagonist. Life Sci. 42,
19411954 (1988).
82. Kikuchi, T. etal. 7-(4-[4-(2,3-dichlorophenyl)
-1-piperazinyl]butyloxy)-3,4-dihydro-2(1H)-
quinolinone (OPC-14597), a new putative
antipsychotic drug with both presynaptic dopamine
autoreceptor agonistic activity and postsynaptic D2
receptor antagonistic activity. J.Pharmacol. Exp. Ther.
274, 329336 (1995).
83. Oshiro, Y. etal. Novel antipsychotic agents with
dopamine autoreceptor agonist properties: synthesis
and pharmacology of 7-[4-(4-phenyl-1-piperazinyl)
butoxy]-3,4-dihydro-2(1H)-quinolinone derivatives.
J.Med. Chem. 41, 658667 (1998).
84. Inoue, T., Domae, M., Yamada, K. & Furukawa, T.
Effects of the novel antipsychotic agent
7-(4-[4-(2,3-dichorophenyl)-1-piperazinyl]
butyloxy)-3,4-dihydro-2(1H)-quinoline (OPC-14597)
on prolactin release from the rat anterior pituitary
gland. J.Pharmacol. Exp. Ther. 277, 137143
(1996).
85. Egger, G., Liang, G., Aparicio, A. & Jones, P.A.
Epigenetics in human disease and prospects for
epigenetic therapy. Nature 429, 457463
(2004).
86. Satistowska-Schroder, E.T., Kerridge, D. & Perry, H.
Echinocandin inhibition of 1,3--D-glucan synthase
from Candida albicans. FEBS Lett. 173, 134138
(1984).
87. Nishi, T. etal. Studies on 2-oxoquinoline derivatives as
blood platelet aggregation inhibitors. I. Alkyl
4-(2-oxo-1,2,3,4-tetrahydro-6-quinolyloxy)butyrates
and related compounds. Chem. Pharm. Bull. 31,
798810 (1983).
88. Tally, F.P. & DeBruin M.F. Development of daptomycin
for Gram-positive infections. J.Antimicrob. Chemother.
46, 523526 (2000).
89. Stock, C, C. & Francis, T.J. The inactivation of the
virus of epidemic influenza by soaps. J.Exp. Med. 71,
661681 (1940).
90. Snipes, W., Person, S., Keller, G., Taylor, W. & Keith, A.
Inactivation of lipid-containing viruses by long-chain
alcohols. Antimicrob. Agents Chemother. 11, 98104
(1977).
91. Sands, J., Auperin, D. & Snipes, W. Extreme sensitivity
of enveloped viruses including herpes-simplex, to
long-chain unsaturated monglycerides and alcohols.
Antimicrob. Agents Chemother. 15, 6773
(1979).
92. Katz, D.H., Marcelletti, J.F., Khalil, M.H., Pope, L.E.
& Katz, L.E. Antiviral activity of 1-docosanol, an
inhibitor of lipid-enveloped viruses including herpes
simplex. Proc. Natl Acad. Sci. USA 88, 1082510829
(1991).
93. Wakeling, A.E., Dukes, M. & Bowler, J. A potent
specific pure antiestrogen with clinical potential.
Cancer Res. 51, 38673873 (1991).
94. Stenoien, D.L. etal. FRAP reveals that mobility of
oestrogen receptor- is ligand- and proteasome-
dependent. Nature Cell Biol. 3, 1523 (2001).
95. Glower, A.J., Noyer, M., Verloes, R., Gobert, J. &
Wulfert, E. UCB L059, a novel anti-convulsant drug:
pharmacological profile in animals. Eur. J.Pharmacol.
222, 193203 (1992).
96. Shinabarger, D. Mechanism of action of the
oxazolidinone antibacterial agents. Expert Opin.
Investig. Drugs 8, 11951202 (1999).
97. Brickner, S.J. Oxazolidinone antibacterial agents.
Curr. Pharm. Des. 2, 175194 (1996).
98. Cuppoletti J. etal. Recombinant and native intestinal
cell ClC-2 Cl
channels are activated by RU-0211.

Gastroenterology 122, A538 (2002).
99. Cuppoletti, J. etal. SPI-0211 activates T84 cell
chloride transport and recombinant human ClC-2
chloride currents. Am. J.Physiol. 287, C1173C1183
(2004).
100. Pea-Mnzenmayer, G. etal. Basolateral localization
of native ClC-2 chloride channels in absorptive
intestinal epithelial cells and basolateral sorting
encoded by a CBS-2 domain di-leucine motif. J.Cell
Sci. 118, 42434252 (2005).
101. Parsons, C.G., Danysz, W. & Quack, G. Memantine is a
clinically well-tolerated N-methyl-d-aspartate (NMDA)
receptor antagonist a review of the preclinical data.
Neuropharmacology 38, 735767 (1999).
102. Gerzon, K. etal. The adamantyl group in medicinal
agents. I. Hypoglycemic N-arylsulfonyl-N-
adamantylureas. J.Med. Chem. 6, 760763 (1963).
103. Bormann, J. Memantine is a potent blocker of
N-methyl-d-aspartate (NMDA) receptor channels. Eur.
J.Pharmacol. 166, 591592 (1989).
104. Platt, F.M., Neises, G.R., Dwek, R.A. & Butters, T.D.
N-butyldeoxynojirimycin is a novel inhibitor of
glycolipid biosynthesis. J.Biol. Chem. 269,
83628365 (1994).
105. Pastores, G.M. & Barnett, N.L. Substrate reduction
therapy: miglustat as a remedy for symptomatic
patients with Gaucher disease type1. Expert Opin.
Investig. Drugs. 12, 273281 (2003).
106. Hu, S. etal. Pancreatic -cell KATP channel activity
and membrane-binding studies with nateglinide: a
comarison with sulfonylureas and repaglinide.
J.Pharmacol. Ther. 293, 444452 (2000).
J.Med. Chem. 31, 20922097 (1988).
J.Med. Chem. 32, 14361441 (1989).
109. Parker, W.B. etal. Purine nucleoside analogues in
development for the treatment of cancer. Curr. Opin.
Investig. Drugs 5, 592596 (2004).
110. Rodriguez, C.O. etal. Mechanisms for T-cell selective
cytotoxicity of arabinosylguanine. Blood 102,
18421848 (2003).
111. Krenitsky, T.A. etal. An enzymatic synthesis of
purine-d-arabinonucleosides. Carbohydr. Res. 97,
139146 (1981).
112. Lambe, C.U. etal. 2-amino-6-methoxypurine
arabinoside: an agent for T-cell malignancies. Cancer
Res. 55, 33523356 (1995).
113. Gandhi, V., Keating, M.J., Bate, G. & Kirkpatrick, P.
Nelarabine. Nature Rev. Drug Discov. 5, 1718 (2006).
114. Lock, E.A. etal. From toxicological problem to
therapeutic use: the discovery of the mode of action
of 2-(2-nitro-4-trifluoromethylbenzoyl)-1,
3-cyclohexanedione (NTBC), its toxicology and
development as a drug. J.Inherit. Metab. Dis. 21,
498506 (1998).
115. Kavana, M. & Moran, G.R. Interaction of
(4-hydroxyphenyl)pyruvate dioxygenase with the
specific inhibitor 2-[2-Nitro-4-(trifluoromethyl)
benzoly]-1,3-cyclohexanedione. Biochemistry 42,
1023810245 (2003).
116. Brownlee, J.M., Johnson-Winters, K., Harrison, D.H. T.
& Moran, G.R. Structure of the ferrous form of
4-(hydroxyphenyl)pyruvate dehydrogenase from
Streptomyces avermitilis in complex with the
therapeutic herbicide, NTBC. Biochemistry 43,
63706377 (2004).
117. Yanagihara, Y., Kasai, H., Kawashima, T. & Shida, T.
Immunopharmacological studies on TBX, a new
antiallergic drug (1). Inhibitory effects on passive
cutaneous anaphylaxis in rats and guinea pigs.
Jpn. J.Pharmacol. 48, 91101 (1988).
118. Gaffney, S.M. Ranolazine, a novel agent for chronic
stable angina. Pharmacotherapy 26, 135142
(2006).
ANALYSI S
nrd_3480_jul11.indd 518 22/06/2011 10:16
Biopharma Dealmakers features a series of business proles from the leading biotech and pharmaceutical
companies responsible for creating tomorrows innovations.
By placing a customized advertising prole in BioPharma Dealmakers, you can communicate your needs,
vision, and unique products and technologies to a global audience of 130,000+ scientic inuencers,
business development professionals and nanciers.

All issues of BioPharma Dealmakers appear in Nature Biotechnology and Nature Reviews Drug Discovery
in both print and online.

Dont miss out on this unique opportunity!
www.nature.com/biopharmadealmakers
Attract attention from leading biopharma companies
and academia, to find that perfect business partner.
12663-25_BioPharmaDealmakers.indd 1 10/06/2011 13:13
Human monoclonal antibodies (mAbs)
are the fastest-growing category of mAb
therapeutics entering clinical study
13
.
Although the technologies developed in the
1970s and early 1980s to produce murine
(rodent-derived) mAbs could be applied to
produce human candidates
4
, few human
mAbs entered clinical development owing
to manufacturing challenges. Murine anti-
bodies are easier to produce, but are limited
by safety issues and diminished efficacy
owing to the immunogenicity of the
mouse-derived protein sequences.
In the mid-1980s, several avenues were
explored to improve the characteristics of
therapeutic mAbs, based in part on the
hypothesis that reducing the extent of,
or eliminating, mouse-derived sequences
would reduce mAb immunogenicity
(BOX 1). One path focused on the develop-
ment of mAbs that contained a combination
of rodent-derived and human-derived
sequences, resulting in chimeric and human-
ized mAbs. These versions constituted the
majority of candidates in clinical study
during the 1990s (FIG. 1), and two-thirds
of the 24 mAbs currently on the market
in the United States are either chimeric
or humanized products.
An alternative path focused on
the generation of human mAbs from
transgenic-mouse technologies and
phage-display technologies. However, patent
disputes impeded broad use of these
methods and contributed to the dearth of
candidates in the clinic during the 1990s.
During the 2000s, human mAbs constituted
45% of the mAb candidates in the clinic
(FIG. 1), and 88 are now in clinical develop-
ment. So far, seven human mAbs have been
approved for marketing in the United States:
adalimumab (Humira; Abbott), panitu-
mumab (Vectibix; Amgen), golimumab
(Simponi; Centocor), canakinumab (Ilaris;
Novartis), ustekinumab (Stelara; Johnson &
Johnson), ofatumumab (Arzerra; Genmab)
and denosumab (Prolia; Amgen). Moreover,
three candidates are undergoing review
by the US Food and Drug Administration
(FDA): raxibacumab and belimumab, both
under development by Human Genome
Sciences, and ipilimumab, under
development by Bristol-Myers Squibb.
To determine trends in the clinical study
of human mAbs, we analysed data for a
total of 147 candidates that entered clinical
study between 1985 and 2008, focusing
primarily on the 131 candidates that entered
studies after 1996 (see BOX 2 for an explana-
tion of data sources and methods). Owing to
the large body of literature on these mAbs,
only limited references to the primary litera-
ture are given. We analysed data for human
mAbs as a single cohort and stratified
the data by clinical indication and source
technology to determine developmental
trends and rates of approval success in the
United States. Our results should inform
the future research and development of
these therapeutics.
Clinical development, 19851996
Human mAb therapeutics first entered the
clinic in the mid-1980s, but only 16 human
mAbs that fit the selection criteria (BOX 2)
entered clinical development during the
12-year period of 19851996. By contrast,
131 human mAbs were first studied in the
clinic during the following 12-year period
(19972008) (FIG. 2). As recombinant DNA
technology was at an early stage of develop-
ment in the 1980s, human mAbs could
be produced through only a few methods
for example, through the generation of
human hybridomas derived from human
lymphocytes and from myeloma cell
lines
57
or from immortalization of primary
human lymphocytes using the EpsteinBarr
virus
8,9
. Although these approaches were
innovative at the time, they proved to be
unreliable, produced insufficient quantities
of antibodies and were vulnerable to con-
tamination
4,10
. An additional limitation was
that the source cells were lymphocytes from
patients, who produced the cells through
natural processes; for ethical reasons, it was
not possible to immunize patients with
an experimental antigen in a controlled
manner and then collect the resulting lym-
phocytes. Early human mAbs evaluated in
the clinic were therefore limited to targets
relevant to infectious diseases (62.5%) and
cancer (37.5%).
OUTLOOK
Development trends for human
monoclonal antibody therapeutics
Aaron L. Nelson, Eugen Dhimolea and Janice M. Reichert
Abstract | Fully human monoclonal antibodies (mAbs) are a promising and rapidly
growing category of targeted therapeutic agents. The first such agents were
developed during the 1980s, but none achieved clinical or commercial success.
Advances in technology to generate the molecules for study in particular,
transgenic mice and yeast or phage display renewed interest in the development
of human mAbs during the 1990s. In 2002, adalimumab became the first human
mAb to be approved by the US Food and Drug Administration (FDA). Since then,
an additional six human mAbs have received FDA approval: panitumumab,
golimumab, canakinumab, ustekinumab, ofatumumab and denosumab. In addition,
3 candidates (raxibacumab, belimumab and ipilimumab) are currently under review
by the FDA, 7 are in Phase III studies and 81 are in either Phase I or II studies. Here,
we analyse data on 147 human mAbs that have entered clinical study to highlight
trends in their development and approval, which may help inform future studies of
this class of therapeutic agents.
PerSPecTIveS
NATUrE rEVIEWS | Drug Discovery VOlUME 9 | OCTOBEr 2010 | 767
nrd_3229_oct10.indd 767 17/9/10 11:50:12
Biopharma Dealmakers features a series of business proles from the leading biotech and pharmaceutical
companies responsible for creating tomorrows innovations.
By placing a customized advertising prole in BioPharma Dealmakers, you can communicate your needs,
vision, and unique products and technologies to a global audience of 130,000+ scientic inuencers,
business development professionals and nanciers.

All issues of BioPharma Dealmakers appear in Nature Biotechnology and Nature Reviews Drug Discovery
in both print and online.

Dont miss out on this unique opportunity!
www.nature.com/biopharmadealmakers
Attract attention from leading biopharma companies
and academia, to find that perfect business partner.
12663-25_BioPharmaDealmakers.indd 1 10/06/2011 13:13
Human monoclonal antibodies (mAbs)
are the fastest-growing category of mAb
therapeutics entering clinical study
13
.
Although the technologies developed in the
1970s and early 1980s to produce murine
(rodent-derived) mAbs could be applied to
produce human candidates
4
, few human
mAbs entered clinical development owing
to manufacturing challenges. Murine anti-
bodies are easier to produce, but are limited
by safety issues and diminished efficacy
owing to the immunogenicity of the
mouse-derived protein sequences.
In the mid-1980s, several avenues were
explored to improve the characteristics of
therapeutic mAbs, based in part on the
hypothesis that reducing the extent of,
or eliminating, mouse-derived sequences
would reduce mAb immunogenicity
(BOX 1). One path focused on the develop-
ment of mAbs that contained a combination
of rodent-derived and human-derived
sequences, resulting in chimeric and human-
ized mAbs. These versions constituted the
majority of candidates in clinical study
during the 1990s (FIG. 1), and two-thirds
of the 24 mAbs currently on the market
in the United States are either chimeric
or humanized products.
An alternative path focused on
the generation of human mAbs from
transgenic-mouse technologies and
phage-display technologies. However, patent
disputes impeded broad use of these
methods and contributed to the dearth of
candidates in the clinic during the 1990s.
During the 2000s, human mAbs constituted
45% of the mAb candidates in the clinic
(FIG. 1), and 88 are now in clinical develop-
ment. So far, seven human mAbs have been
approved for marketing in the United States:
adalimumab (Humira; Abbott), panitu-
mumab (Vectibix; Amgen), golimumab
(Simponi; Centocor), canakinumab (Ilaris;
Novartis), ustekinumab (Stelara; Johnson &
Johnson), ofatumumab (Arzerra; Genmab)
and denosumab (Prolia; Amgen). Moreover,
three candidates are undergoing review
(FDA): raxibacumab and belimumab, both
under development by Human Genome
Sciences, and ipilimumab, under
To determine trends in the clinical study
of human mAbs, we analysed data for a
total of 147 candidates that entered clinical
study between 1985 and 2008, focusing
primarily on the 131 candidates that entered
studies after 1996 (see BOX 2 for an explana-
tion of data sources and methods). Owing to
the large body of literature on these mAbs,
only limited references to the primary litera-
ture are given. We analysed data for human
mAbs as a single cohort and stratified
the data by clinical indication and source
technology to determine developmental
trends and rates of approval success in the
United States. Our results should inform
the future research and development of
these therapeutics.
Human mAb therapeutics first entered the
clinic in the mid-1980s, but only 16 human
mAbs that fit the selection criteria (BOX 2)
entered clinical development during the
12-year period of 19851996. By contrast,
131 human mAbs were first studied in the
clinic during the following 12-year period
(19972008) (FIG. 2). As recombinant DNA
technology was at an early stage of develop-
ment in the 1980s, human mAbs could
be produced through only a few methods
for example, through the generation of
human hybridomas derived from human
lymphocytes and from myeloma cell
lines
57
or from immortalization of primary
human lymphocytes using the EpsteinBarr
virus
8,9
. Although these approaches were
innovative at the time, they proved to be
unreliable, produced insufficient quantities
of antibodies and were vulnerable to con-
tamination
4,10
. An additional limitation was
that the source cells were lymphocytes from
patients, who produced the cells through
natural processes; for ethical reasons, it was
not possible to immunize patients with
an experimental antigen in a controlled
manner and then collect the resulting lym-
phocytes. Early human mAbs evaluated in
the clinic were therefore limited to targets
relevant to infectious diseases (62.5%) and
cancer (37.5%).
OUTLOOK
Development trends for human
monoclonal antibody therapeutics
Aaron L. Nelson, Eugen Dhimolea and Janice M. Reichert
Abstract | Fully human monoclonal antibodies (mAbs) are a promising and rapidly
growing category of targeted therapeutic agents. The first such agents were
developed during the 1980s, but none achieved clinical or commercial success.
Advances in technology to generate the molecules for study in particular,
transgenic mice and yeast or phage display renewed interest in the development
of human mAbs during the 1990s. In 2002, adalimumab became the first human
mAb to be approved by the US Food and Drug Administration (FDA). Since then,
an additional six human mAbs have received FDA approval: panitumumab,
golimumab, canakinumab, ustekinumab, ofatumumab and denosumab. In addition,
3 candidates (raxibacumab, belimumab and ipilimumab) are currently under review
by the FDA, 7 are in Phase III studies and 81 are in either Phase I or II studies. Here,
we analyse data on 147 human mAbs that have entered clinical study to highlight
trends in their development and approval, which may help inform future studies of
this class of therapeutic agents.
PerSPecTIveS
nrd_3229_oct10.indd 767 17/9/10 11:50:12
First published in Nature Reviews Drug Discovery 9, 767-774 (October 2010) | doi:10.1038/nrd3229
m
A
b

p
i
p
e
l
i
n
e

(
%
)
50
45
40
35
30
25
20
15
10
5
0
Humanized Human Chimeric Murine
30
7
13.5
9
45
39
11.5
45 19901999
20002008
Approval was based on a single Phase III
trial
20
, as well as on smaller proof-of-concept
studies
21
. CAPS are rare, with the number
of patients with CAPS living in the United
States estimated to be in the hundreds or low
thousands. In October 2009, canakinumab
was approved in the European Union for
patients with CAPS as young as 4 years old.
Canakinumab is currently in early-stage
studies for the treatment of other disorders,
including rheumatoid arthritis, gout and
diabetes mellitus.
Ustekinumab, another UltiMab-derived
product, targets the p40 subunit shared by
Il-12 and Il-23. The mAb was approved by
the FDA in September 2009 as a treatment
for plaque psoriasis. Two Phase III studies in
patients with moderate to severe plaque
psoriasis have been completed, as well as a
third Phase III study comparing ustekinumab
with etanercept (Enbrel; Amgen/Pfizer),
a fusion protein that targets TNF, in the same
patient population
22
. In January 2009, the
European Commission approved ustekinu-
mab for treating moderate to severe plaque
psoriasis in adults. Ustekinumab is also
currently in Phase II studies as a treatment
for sarcoidosis, and in Phase III studies as a
treatment for palmoplantar pustular psoriasis,
palmoplantar pustulosis or psoriatic arthritis.
Ofatumumab is a CD20-specific mAb
generated from the UltiMab platform.
It targets a CD20 epitope that is distinct
from the epitope targeted by rituximab
(rituxan/MabThera; Genentech/Biogen
Idec/roche), the pioneering CD20-specific
chimeric mAb. rituximab was approved
for the treatment of non-Hodgkins lym-
phoma in 1997, and subsequently has also
received regulatory approval for the treat-
ment of rheumatoid arthritis and chronic
lymphocytic leukaemia
23,24
. Ofatumumab
was approved by the FDA in October 2009,
and given a conditional approval by the
European Commission in April 2010, for
the treatment of chronic lymphocytic leu-
kaemia that is refractory to the humanized
mAb alemtuzumab (Campath; Genzyme)
and the nucleoside analogue fludarabine.
Ofatumumab is under Phase III evaluation
in patients with non-Hodgkins lymphoma
and in patients with rheumatoid arthritis
25
.
Denosumab, a mAb specific for receptor
activator of nuclear factor-B ligand
(rANKl), was approved by the FDA in
June 2010 for the treatment of postmeno-
pausal osteoporosis (PMO) in women.
Trials have also been conducted to support
a prevention indication for PMO, as well as
for the treatment and prevention of bone loss
in patients undergoing hormone ablation
therapy for prostate or breast cancer
26
.
Denosumab was approved in Europe for the
treatment of PMO and of bone loss in patients
with prostate cancer undergoing hormone
ablation therapy. It is also undergoing regu-
latory review in Switzerland, Australia and
Canada for one or more of these indications.
Although not yet approved, three human
mAbs raxibacumab, belimumab and
ipilimumab are currently undergoing
review by the FDA. Human Genome
Sciences is the sponsor of both raxibacumab
and belimumab, and Bristol-Myers Squibb
is sponsoring ipilimumab. raxibacumab
binds Bacillus anthracis protective antigen
and has been developed as a treatment
for inhalation anthrax
27
. Human Genome
Sciences initiated the delivery of 20,000
doses of raxibacumab to the US Strategic
National Stockpile for emergency use
under a contract with the US Biomedical
Advanced research and Development
Authority, and an additional 45,000 doses
were ordered in July 2009.
Belimumab is a human mAb specific for
B lymphocyte stimulator, and was identified
through use of phage-display-based tech-
nologies in collaboration with Cambridge
Antibody Technology. GlaxoSmithKline
and Human Genome Sciences submitted
marketing applications to both the FDA and
the European Medicines Agency in June
2010 for the use of belimumab in systemic
lupus erythematosus. This submission is
based primarily on clinical and biomarker
improvements in two pivotal Phase III
trials in systemic lupus erythematosus
BlISS-52 and BlISS-76, which collectively
involved 1,684 patients with this disease
globally
28
as well as favourable post-hoc
analyses of Phase II studies
29
. If the applica-
tion is successful, belimumab will be the first
new therapeutic approved for systemic lupus
erythematosus in 50 years. GlaxoSmithKline
and Human Genome Sciences are further
Figure 1 | Percentage of four types of mAbs in
clinical development during the periods
19901999 and 20002008. Monoclonal anti-
bodies (mAbs) that entered clinical study spon-
sored by commercial firms between 1990 and
1999 and between 2000 and 2008 were classified
according to their sequence source: murine only,
chimeric (murine variable regions and human
constant regions), humanized (human with
murine complementarity-determining regions),
and human only. These data demonstrate the
substantial increase in the clinical study of human
mAbs in the 2000s, a trend towards reduced use
of humanization and chimeric candidates, and a
dramatic reduction in the number of murine
mAbs in clinical development in the 2000s.

Box 2 | Analysis criteria
Since it was founded in 1976, the Tufts Center for the Study of Drug Development has collected
data on the clinical study and approval of therapeutics and vaccines. Data for monoclonal
antibodies (mAbs) were collected by surveying pharmaceutical and biotechnology firms,
from public documents and from commercially available databases (IDdb3, IMS R&D Focus
and PharmaProjects). Data were updated with all changes that were noted until June 2010.
The data set comprises a total of 147 human mAbs that entered clinical study sponsored by
commercial firms between January 1985 and December 2008, 131 of which entered study
between 1997 and 2008. The status of the candidates was as follows: 88 were in clinical studies
and not yet approved in any country (30 in Phase I, 51 in Phase II and 7 in Phase III); 3 were under
regulatory review by the US Food and Drug Administration; 7 were approved in the United States;
and 49 were discontinued. Candidates in Phase I or II were assigned to Phase II, and products
in Phase II or III were assigned to Phase III. The human mAb data were compared with data for
humanized mAbs that entered clinical study between 1988 and 2008 (n = 167) and between 1997
and 2008 (n = 133).
Approval success calculations were based on data for candidates with known fates (market
approval in the United States or discontinuation of all clinical studies). Percentage completion
was defined as the percentage of candidates with a known fate in a given cohort. Clinical-phase
transition probabilities were calculated as follows: the number of candidates that successfully
completed a given phase was divided by the difference between the number of candidates
that entered the phase and those that were still in the phase at the time of the calculation.
Transitions occurring between phases of clinical studies conducted worldwide were included.
PersPecti ves
nrd_3229_oct10.indd 769 17/9/10 11:50:12
Fifteen out of the 16 early human mAb
candidates were terminated during clinical
development. One candidate nebacumab
(Centoxin; Centocor) was approved for
marketing, but the product was subsequently
withdrawn. Nebacumab was a human
hybridoma-derived endotoxin-specific
immunoglobulin M (IgM) mAb. It was
approved for the treatment of sepsis or
Gram-negative bacteraemia
11
and was the
first human therapeutic mAb to be reviewed
by a regulatory agency. Statistically significant
benefits that were observed post-hoc in sub-
populations in a multi-centre, randomized,
double-blind, placebo-controlled clinical
trial
12
supported marketing approval in
several European countries and in New
Zealand. A marketing application was submit-
ted in the United States, but a second study
required by the FDA
13,14
, the CHESS trial, was
terminated early when an interim analysis
found a non-statistically significant increase in
mortality in patients without Gram-negative
bacteraemia treated with the antibody
15
.
Centocor voluntarily withdrew the product
and suspended further development.
In the late 1990s, human mAb therapeutics
derived from transgenic-mouse or phage-
display technologies first entered clinical
development. Between 1997 and 2008, a
total of 131 human mAbs entered clinical
study, at a rate of least 11 per year between
2001 and 2008 (FIG. 2). Of the 131 candidates,
88 were in active clinical development,
with 7 in Phase III studies, 51 in Phase II
and 30 in Phase I. A total of 7 were approved
for marketing by the FDA, 3 are undergoing
review by the FDA and the clinical study of
33 was discontinued.
Approved human mAbs
Seven human mAbs have been approved in
the United States and in the European Union
(TABLE 1). The first product (adalimumab)
was approved by the FDA in 2002, with the
second following in 2006. Notably, a total
of four human mAbs were approved by the
FDA in 2009.
Adalimumab is specific for tumour
necrosis factor (TNF) and was the first
human mAb approved by the FDA. The
product, which was developed using
phage-display technology from Cambridge
Antibody Technology, was approved in
December 2002 as a treatment for adult
patients with moderately to severely active
rheumatoid arthritis. Adalimumab was
subsequently approved by the FDA for the
following indications: psoriatic arthritis
(in 2005), ankylosing spondylitis (in 2006)
and Crohns disease (in 2007), as well as for
juvenile idiopathic arthritis and chronic
plaque psoriasis (both in 2008). Adalimumab
is also approved for the treatment of these
diseases in the European Union. According
to the manufacturer, adalimumab generated
US$4.5 billion in global sales in 2008 (REF. 16).
Panitumumab is a human mAb that is
specific for epidermal growth factor receptor
(EGFr) and was discovered using Abgenixs
XenoMouse technology. The product was
approved by the FDA in September 2006
for EGFr-expressing refractory metastatic
colorectal carcinoma. The clinical develop-
ment programme comprised a total of 15
studies initiated before approval, including
10 Phase I studies
17
. The product was given
accelerated approval based on results from
one randomized, controlled trial involving
463 patients who showed prolongation
of the time to disease progression from
60 days to 97 days, but no impact on overall
survival
18
. In December 2007, the European
Commission granted conditional marketing
approval for panitumumab as a treatment
for EGFr-expressing metastatic colon
cancer. Panitumumab generated global sales
of $153 million in 2008.
Golimumab is a TNF-specific IgG1 mAb
that was approved in April 2009 by the FDA
for the treatment of rheumatoid arthritis,
psoriatic arthritis and ankylosing spondylitis.
The mAb was generated using Medarexs
UltiMab transgenic mouse platform. Although
both adalimumab and golimumab are
human TNF-specific mAbs, golimumab has
a once per month subcutaneous dosing
regimen, whereas adalimumab is adminis-
tered every other week
19
. In October 2009,
golimumab was approved in the European
Union as a once per month, subcutaneous
therapy for the treatment of moderate to
severe, active rheumatoid arthritis, of active
and progressive psoriatic arthritis, and of
severe, active ankylosing spondylitis.
Canakinumab, an interleukin-1
(Il-1)-specific IgG1 mAb derived from the
UltiMab platform technology was approved
by the FDA in June 2009 as a treatment
for cryopyrin-associated periodic syndromes
(CAPS), which include rare genetic fever
disorders such as MuckleWells syndrome.

Box 1 | Immunogenicity of human monoclonal antibodies
The immunogenicity of therapeutic proteins, including monoclonal antibodies (mAbs), affects
the safety and efficacy of these products
40
. Immune responses to therapeutic mAbs are
undesirable as they can neutralize the action of therapeutic mAbs
41
, and hypersensitivity
can result in morbidity and mortality
42
. For example, development of antibodies against
adalimumab (Humira; Abbott) has been associated with lower serum drug levels and poor
clinical response
43,44
.
Development of human mAbs was based on the hypothesis that they would prove to be less
immunogenic than chimeric or humanized mAbs, both of which contain some murine-derived
protein sequences. In general, eliminating rodent sequences reduces the frequency of
mAb-targeted immune responses and hypersensitivity reactions
45
. For example, only 1%
of patients treated with panitumumab (Vectibix; Amgen) tested positive for neutralizing
antibodies
46,47
. By contrast, the presence of pretreatment serum autoantibodies (approximately
22% of patients) against the chimeric mAb cetuximab (Erbitux; Bristol-Myers Squibb/Merck/
ImClone Systems) was significantly associated with patient hypersensitivity
48
. However, the
immunogenicity of specific mAb candidates cannot be predicted based only on the amount
of non-human sequence in the molecule. This is because various other factors can affect
immunogenicity rates for example, the type of disease being treated
43,49
or the concomitant
administration of immunosuppresive agents
50
. In addition, immunogenicity rates can vary
between studies. Different methods in different studies, such as enzyme-linked immunosorbent
assay or surface plasmon resonance, may be used to quantify drug-specific antibodies
51
, and the
results reported will depend on the sensitivity of the assay
52
.
Several lines of thought suggest that it is unreasonable to expect human mAb therapeutics
to have immunogenicity rates of zero. For example, humans are diverse in genotype and
phenotype, and immunoglobulin G allotypes differ within and between populations
53
.
In addition, the ability of the human immune system to generate natural anti-idiotypic
antibodies is well documented
54
; the presence of such antibodies in polyclonal intravenous
immunoglobulin preparations may contribute to the efficacy of the products
55
. Various methods
to reduce the immunogenicity of human mAbs have been suggested, such as production of
allotypical variants of mAb products to match the specific immunoglobulin gene segment
alleles found in the genomes of distinct patient populations, or the use of protein engineering
on the complementarity-determining region of the mAb
53,56
. Although the causes of, and
potential solutions to, immunogenicity of human mAbs are still being investigated, it is clear
that immunogenicity must be included as part of the overall evaluation of the risk to benefit
ratio for patients
57
.
PersPecti ves
768 | OCTOBEr 2010 | VOlUME 9 www.nature.com/reviews/drugdisc
nrd_3229_oct10.indd 768 17/9/10 11:50:12
m
A
b

p
i
p
e
l
i
n
e

(
%
)
50
45
40
35
30
25
20
15
10
5
0
Humanized Human Chimeric Murine
30
7
13.5
9
45
39
11.5
45 19901999
20002008
Approval was based on a single Phase III
trial
20
, as well as on smaller proof-of-concept
studies
21
. CAPS are rare, with the number
of patients with CAPS living in the United
States estimated to be in the hundreds or low
thousands. In October 2009, canakinumab
was approved in the European Union for
patients with CAPS as young as 4 years old.
Canakinumab is currently in early-stage
studies for the treatment of other disorders,
including rheumatoid arthritis, gout and
diabetes mellitus.
Ustekinumab, another UltiMab-derived
product, targets the p40 subunit shared by
Il-12 and Il-23. The mAb was approved by
the FDA in September 2009 as a treatment
for plaque psoriasis. Two Phase III studies in
patients with moderate to severe plaque
psoriasis have been completed, as well as a
third Phase III study comparing ustekinumab
with etanercept (Enbrel; Amgen/Pfizer),
a fusion protein that targets TNF, in the same
patient population
22
. In January 2009, the
European Commission approved ustekinu-
mab for treating moderate to severe plaque
psoriasis in adults. Ustekinumab is also
currently in Phase II studies as a treatment
for sarcoidosis, and in Phase III studies as a
treatment for palmoplantar pustular psoriasis,
palmoplantar pustulosis or psoriatic arthritis.
Ofatumumab is a CD20-specific mAb
generated from the UltiMab platform.
It targets a CD20 epitope that is distinct
from the epitope targeted by rituximab
(rituxan/MabThera; Genentech/Biogen
Idec/roche), the pioneering CD20-specific
chimeric mAb. rituximab was approved
for the treatment of non-Hodgkins lym-
phoma in 1997, and subsequently has also
received regulatory approval for the treat-
ment of rheumatoid arthritis and chronic
lymphocytic leukaemia
23,24
. Ofatumumab
was approved by the FDA in October 2009,
and given a conditional approval by the
European Commission in April 2010, for
the treatment of chronic lymphocytic leu-
kaemia that is refractory to the humanized
mAb alemtuzumab (Campath; Genzyme)
and the nucleoside analogue fludarabine.
Ofatumumab is under Phase III evaluation
in patients with non-Hodgkins lymphoma
and in patients with rheumatoid arthritis
25
.
Denosumab, a mAb specific for receptor
activator of nuclear factor-B ligand
(rANKl), was approved by the FDA in
June 2010 for the treatment of postmeno-
pausal osteoporosis (PMO) in women.
Trials have also been conducted to support
a prevention indication for PMO, as well as
for the treatment and prevention of bone loss
in patients undergoing hormone ablation
therapy for prostate or breast cancer
26
.
Denosumab was approved in Europe for the
treatment of PMO and of bone loss in patients
with prostate cancer undergoing hormone
ablation therapy. It is also undergoing regu-
latory review in Switzerland, Australia and
Canada for one or more of these indications.
Although not yet approved, three human
mAbs raxibacumab, belimumab and
ipilimumab are currently undergoing
review by the FDA. Human Genome
Sciences is the sponsor of both raxibacumab
and belimumab, and Bristol-Myers Squibb
is sponsoring ipilimumab. raxibacumab
binds Bacillus anthracis protective antigen
and has been developed as a treatment
for inhalation anthrax
27
. Human Genome
Sciences initiated the delivery of 20,000
doses of raxibacumab to the US Strategic
National Stockpile for emergency use
under a contract with the US Biomedical
Advanced research and Development
Authority, and an additional 45,000 doses
were ordered in July 2009.
Belimumab is a human mAb specific for
B lymphocyte stimulator, and was identified
through use of phage-display-based tech-
nologies in collaboration with Cambridge
Antibody Technology. GlaxoSmithKline
and Human Genome Sciences submitted
marketing applications to both the FDA and
the European Medicines Agency in June
2010 for the use of belimumab in systemic
lupus erythematosus. This submission is
based primarily on clinical and biomarker
improvements in two pivotal Phase III
trials in systemic lupus erythematosus
BlISS-52 and BlISS-76, which collectively
involved 1,684 patients with this disease
globally
28
as well as favourable post-hoc
analyses of Phase II studies
29
. If the applica-
tion is successful, belimumab will be the first
new therapeutic approved for systemic lupus
erythematosus in 50 years. GlaxoSmithKline
and Human Genome Sciences are further
Figure 1 | Percentage of four types of mAbs in
clinical development during the periods
19901999 and 20002008. Monoclonal anti-
bodies (mAbs) that entered clinical study spon-
sored by commercial firms between 1990 and
1999 and between 2000 and 2008 were classified
according to their sequence source: murine only,
chimeric (murine variable regions and human
constant regions), humanized (human with
murine complementarity-determining regions),
and human only. These data demonstrate the
substantial increase in the clinical study of human
mAbs in the 2000s, a trend towards reduced use
of humanization and chimeric candidates, and a
dramatic reduction in the number of murine
mAbs in clinical development in the 2000s.

Box 2 | Analysis criteria
Since it was founded in 1976, the Tufts Center for the Study of Drug Development has collected
data on the clinical study and approval of therapeutics and vaccines. Data for monoclonal
antibodies (mAbs) were collected by surveying pharmaceutical and biotechnology firms,
from public documents and from commercially available databases (IDdb3, IMS R&D Focus
and PharmaProjects). Data were updated with all changes that were noted until June 2010.
The data set comprises a total of 147 human mAbs that entered clinical study sponsored by
commercial firms between January 1985 and December 2008, 131 of which entered study
between 1997 and 2008. The status of the candidates was as follows: 88 were in clinical studies
and not yet approved in any country (30 in Phase I, 51 in Phase II and 7 in Phase III); 3 were under
regulatory review by the US Food and Drug Administration; 7 were approved in the United States;
and 49 were discontinued. Candidates in Phase I or II were assigned to Phase II, and products
in Phase II or III were assigned to Phase III. The human mAb data were compared with data for
humanized mAbs that entered clinical study between 1988 and 2008 (n = 167) and between 1997
and 2008 (n = 133).
Approval success calculations were based on data for candidates with known fates (market
approval in the United States or discontinuation of all clinical studies). Percentage completion
was defined as the percentage of candidates with a known fate in a given cohort. Clinical-phase
transition probabilities were calculated as follows: the number of candidates that successfully
completed a given phase was divided by the difference between the number of candidates
that entered the phase and those that were still in the phase at the time of the calculation.
Transitions occurring between phases of clinical studies conducted worldwide were included.
PersPecti ves
nrd_3229_oct10.indd 769 17/9/10 11:50:12
Fifteen out of the 16 early human mAb
candidates were terminated during clinical
development. One candidate nebacumab
(Centoxin; Centocor) was approved for
marketing, but the product was subsequently
withdrawn. Nebacumab was a human
hybridoma-derived endotoxin-specific
immunoglobulin M (IgM) mAb. It was
approved for the treatment of sepsis or
Gram-negative bacteraemia
11
and was the
first human therapeutic mAb to be reviewed
by a regulatory agency. Statistically significant
benefits that were observed post-hoc in sub-
populations in a multi-centre, randomized,
double-blind, placebo-controlled clinical
trial
12
supported marketing approval in
several European countries and in New
Zealand. A marketing application was submit-
ted in the United States, but a second study
required by the FDA
13,14
, the CHESS trial, was
terminated early when an interim analysis
found a non-statistically significant increase in
mortality in patients without Gram-negative
bacteraemia treated with the antibody
15
.
Centocor voluntarily withdrew the product
and suspended further development.
In the late 1990s, human mAb therapeutics
derived from transgenic-mouse or phage-
display technologies first entered clinical
development. Between 1997 and 2008, a
total of 131 human mAbs entered clinical
study, at a rate of least 11 per year between
2001 and 2008 (FIG. 2). Of the 131 candidates,
88 were in active clinical development,
with 7 in Phase III studies, 51 in Phase II
and 30 in Phase I. A total of 7 were approved
for marketing by the FDA, 3 are undergoing
review by the FDA and the clinical study of
33 was discontinued.
Approved human mAbs
Seven human mAbs have been approved in
the United States and in the European Union
(TABLE 1). The first product (adalimumab)
was approved by the FDA in 2002, with the
second following in 2006. Notably, a total
of four human mAbs were approved by the
FDA in 2009.
Adalimumab is specific for tumour
necrosis factor (TNF) and was the first
human mAb approved by the FDA. The
product, which was developed using
phage-display technology from Cambridge
Antibody Technology, was approved in
December 2002 as a treatment for adult
patients with moderately to severely active
rheumatoid arthritis. Adalimumab was
subsequently approved by the FDA for the
following indications: psoriatic arthritis
(in 2005), ankylosing spondylitis (in 2006)
and Crohns disease (in 2007), as well as for
juvenile idiopathic arthritis and chronic
plaque psoriasis (both in 2008). Adalimumab
is also approved for the treatment of these
diseases in the European Union. According
to the manufacturer, adalimumab generated
US$4.5 billion in global sales in 2008 (REF. 16).
Panitumumab is a human mAb that is
specific for epidermal growth factor receptor
(EGFr) and was discovered using Abgenixs
XenoMouse technology. The product was
approved by the FDA in September 2006
for EGFr-expressing refractory metastatic
colorectal carcinoma. The clinical develop-
ment programme comprised a total of 15
studies initiated before approval, including
10 Phase I studies
17
. The product was given
accelerated approval based on results from
one randomized, controlled trial involving
463 patients who showed prolongation
of the time to disease progression from
60 days to 97 days, but no impact on overall
survival
18
. In December 2007, the European
Commission granted conditional marketing
approval for panitumumab as a treatment
for EGFr-expressing metastatic colon
cancer. Panitumumab generated global sales
of $153 million in 2008.
Golimumab is a TNF-specific IgG1 mAb
that was approved in April 2009 by the FDA
for the treatment of rheumatoid arthritis,
psoriatic arthritis and ankylosing spondylitis.
The mAb was generated using Medarexs
UltiMab transgenic mouse platform. Although
both adalimumab and golimumab are
human TNF-specific mAbs, golimumab has
a once per month subcutaneous dosing
regimen, whereas adalimumab is adminis-
tered every other week
19
. In October 2009,
golimumab was approved in the European
Union as a once per month, subcutaneous
therapy for the treatment of moderate to
severe, active rheumatoid arthritis, of active
and progressive psoriatic arthritis, and of
severe, active ankylosing spondylitis.
Canakinumab, an interleukin-1
(Il-1)-specific IgG1 mAb derived from the
UltiMab platform technology was approved
by the FDA in June 2009 as a treatment
for cryopyrin-associated periodic syndromes
(CAPS), which include rare genetic fever
disorders such as MuckleWells syndrome.

Box 1 | Immunogenicity of human monoclonal antibodies
The immunogenicity of therapeutic proteins, including monoclonal antibodies (mAbs), affects
the safety and efficacy of these products
40
. Immune responses to therapeutic mAbs are
undesirable as they can neutralize the action of therapeutic mAbs
41
, and hypersensitivity
can result in morbidity and mortality
42
. For example, development of antibodies against
adalimumab (Humira; Abbott) has been associated with lower serum drug levels and poor
clinical response
43,44
.
Development of human mAbs was based on the hypothesis that they would prove to be less
immunogenic than chimeric or humanized mAbs, both of which contain some murine-derived
protein sequences. In general, eliminating rodent sequences reduces the frequency of
mAb-targeted immune responses and hypersensitivity reactions
45
. For example, only 1%
of patients treated with panitumumab (Vectibix; Amgen) tested positive for neutralizing
antibodies
46,47
. By contrast, the presence of pretreatment serum autoantibodies (approximately
22% of patients) against the chimeric mAb cetuximab (Erbitux; Bristol-Myers Squibb/Merck/
ImClone Systems) was significantly associated with patient hypersensitivity
48
. However, the
immunogenicity of specific mAb candidates cannot be predicted based only on the amount
of non-human sequence in the molecule. This is because various other factors can affect
immunogenicity rates for example, the type of disease being treated
43,49
or the concomitant
administration of immunosuppresive agents
50
. In addition, immunogenicity rates can vary
between studies. Different methods in different studies, such as enzyme-linked immunosorbent
assay or surface plasmon resonance, may be used to quantify drug-specific antibodies
51
, and the
results reported will depend on the sensitivity of the assay
52
.
Several lines of thought suggest that it is unreasonable to expect human mAb therapeutics
to have immunogenicity rates of zero. For example, humans are diverse in genotype and
phenotype, and immunoglobulin G allotypes differ within and between populations
53
.
In addition, the ability of the human immune system to generate natural anti-idiotypic
antibodies is well documented
54
; the presence of such antibodies in polyclonal intravenous
immunoglobulin preparations may contribute to the efficacy of the products
55
. Various methods
to reduce the immunogenicity of human mAbs have been suggested, such as production of
allotypical variants of mAb products to match the specific immunoglobulin gene segment
alleles found in the genomes of distinct patient populations, or the use of protein engineering
on the complementarity-determining region of the mAb
53,56
. Although the causes of, and
potential solutions to, immunogenicity of human mAbs are still being investigated, it is clear
that immunogenicity must be included as part of the overall evaluation of the risk to benefit
ratio for patients
57
.
PersPecti ves
nrd_3229_oct10.indd 768 17/9/10 11:50:12
overall human mAb cohort. This calculation
was based on data for only 12 molecules for
which definite fates are known, including
adalimumab, golimumab, canakinumab,
ustekinumab and 8 terminated candidates.
Of the 24 human immunomodulatory mAb
candidates currently in clinical development,
8 are in Phase I, 13 are in Phase II, 2 are in
Phase III and belimumab is undergoing
regulatory review in the United States and
in the European Union.
An additional 36 human mAbs that
entered clinical study after 1996 were
studied for non-traditional indications
that is, disorders that are not cancer or
immunological in nature. Half of these were
treatments for infectious diseases, with a
focus on nosocomial, anthrax and chronic
viral infections
31
. The remaining 18 candi-
dates were studied for conditions such as
osteoporosis, respiratory disorders, muscular
dystrophy, Alzheimers disease and pain.
Of the 36 candidates, 1 (denosumab) has
been approved, 1 (raxibacumab) is under-
going review by the FDA (TABLE 1) and 14
were terminated. The current cumulative
success rate for mAbs studied as treatments
for these non-traditional indications is 6.6%,
which will rise to 12.5% if raxibacumab is
approved for inhalation anthrax. The 20
candidates in clinical development were all
in either Phase I or II studies.
Molecular targets
The target of a therapeutic antibody is a
major determinant of its efficacy and safety
profile. Antigenic targets were identified
for 125 of the 131 human mAbs (TABLE 2).
These mAbs targeted a total of 89 unique
antigens, and only 22 antigens were targeted
by two or more human mAbs. Seven antigens
CTlA4, EGFr, fibronectin, insulin-like
growth factor 1 receptor (IGF1r), trans-
forming growth factor- (TGF), TNF and
TNF-related apoptosis-inducing ligand
receptor 2 (TrAIlr2) were targeted by
more than two human mAbs. Of the 125
mAbs, 55 (44%) targeted antigens that are
relevant to antineoplastic diseases, 36 (29%)
targeted antigens relevant to immunological
diseases and 17 (14%) targeted antigens
relevant to infectious diseases.
Of the 55 anticancer candidates with
known targets, 19 mAbs (33%) were
specific for only 5 targets: IGF1r (6 mAbs),
TrAIlr2 (4 mAbs), EGFr (3 mAbs),
CTlA4 (3 mAbs) and fibronectin (3 mAbs).
Of these, only EGFr was among the ten
most frequently targeted antigens for all
anticancer mAbs studied in the clinic from
1980 to 2005 (REF. 32), suggesting that devel-
opers of human mAbs may be focusing on
novel therapeutic strategies. Nevertheless,
at least some human mAbs share oncology
targets with therapeutic mAbs approved by
the FDA, including EGFr (target of panitu-
mumab and cetuximab) and CD20 (target of
rituximab, ibritumomab tiuxetan (Zevalin;
Spectrum Pharmaceuticals), iodine-131
tositumomab (Bexxar; GlaxoSmithKline)
and ofatumumab). Most of the anti-
neoplastic human mAbs target cell-surface
molecules; only two were known to target
soluble factors (hepatocyte growth factor
and platelet-derived growth factor).
By contrast, 24 out of the 36 immuno-
modulatory mAbs with known targets were
raised against cytokines and serum factors.
Interleukins constitute the largest target
group; targeted antigens include Il-1, Il-6,
Il-8, Il-12, Il-13, Il-15, Il-18, Il-17A and
Il-20. Ten out of the 36 mAbs target markers
of leukocyte activity and differentiation,
and are in early-phase clinical trials. Two
immunomodulatory human mAbs with
targets that are unique compared with those
of marketed mAbs are currently in Phase II
studies: briakinumab (developed by Abbott),
which targets the p40 subunit common to
Il-12 and Il-23, and AIN457 (developed by
Novartis), which is an Il-17A-specific mAb.
The 18 human mAbs intended for the
treatment of infectious diseases are a highly
heterogeneous group. Ten are directed
against targets implicated in viral infections,
including HIV (six mAbs), viral hepatitis
(three mAbs) and rabies (one mAb); several
of these candidates are cocktails of more
than one human mAb. Six human mAbs
target bacterial antigens, of which four are
bacterial cytotoxins, such as Clostridium
difficile enterotoxins (MK-3415A; developed
by Merck and Co.) and anthrax protective
antigen (raxibacumab), and three target
cellular features of Pseudomonas aeruginosa
Table 1 | Human mAbs approved or under FDA review*
Human mAb
(trade name;
company name)
Description indication of first
us approval
FDA
designations
Date of first us
(eu) approval
Adalimumab
(Humira; Abbott)
TNF-specific,
IgG1
rheumatoid arthritis S 31 Dec 2002
(8 Sep 2003)
Panitumumab
(vectibix; Amgen)
eGFr-specific,
IgG2
colorectal cancer P, FT, AA 27 Sep 2006
(3 Dec 2007)
Golimumab
(Simponi;
centocor)
TNF-specific,
IgG1
rheumatoid
arthritis, psoriatic
arthritis, ankylosing
spondylitis
S 24 Apr 2009
(1 Oct 2009)
canakinumab
(Ilaris; Novartis)
IL-1-specific,
IgG1
cryopyrin-
associated periodic
syndromes
P, O 18 Jun 2009
(23 Oct 2009)
Ustekinumab
(Stelara; Johnson
& Johnson)
IL-12/IL-23
p40-specific,
IgG1
Plaque psoriasis S 25 Sep 2009
(16 Jan 2009)
Ofatumumab
(Arzerra;
Genmab)
cD20-specific,
IgG1
chronic lymphocytic
leukaemia
P, FT 26 Oct 2009
(19 Apr 2010)
Denosumab
(Prolia; Amgen)
rANKL-
specific, IgG2
Treatment of
postmenopausal
osteoporosis

S 1 Jun 2010
(26 May 2010)
raxibacumab PA-specific,
IgG1
Inhalation anthrax P, FT, O Under review by
the FDA
Belimumab B lymphocyte
stimulator-
specific, IgG1
Systemic lupus
erythematosus
P, FT Under review by
the FDA and the
eMA
Ipilimumab cTLA4-specific,
IgG1
Metastatic
melanoma
P, FT, O Under review by
the FDA and the
eMA
AA, accelerated approval; cTLA, cytotoxic T lymphocyte-associated antigen; eGFr, epidermal growth
factor receptor; eMA, european Medicines Agency; eU, european Union; FDA, US Food and Drug
Administration; FT, FDA fast track drug; Ig, immunoglobulin; IL, interleukin; mAb, monoclonal antibody;
O, FDA orphan drug; P, priority review; PA, Bacillus anthracis protective antigen; rANKL, receptor for
activation of nuclear factor-B ligand; S, standard review; TNF, tumour necrosis factor. *As of June 2010.
Also approved in europe for the treatment of bone loss in patients with prostate cancer undergoing
hormone ablation therapy.
PersPecti ves
nrd_3229_oct10.indd 771 17/9/10 11:50:14
N
u
m
b
e
r

o
f

c
l
i
n
i
c
a
l

c
a
n
d
i
d
a
t
e
s
150
90
100
110
120
130
140
80
70
60
50
40
30
20
10
0
1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007
All human mAbs
Antineoplastic only
Immunomodulatory only
Anti-infective only
Other indications
sponsoring a series of Phase II studies of
belimumab in rheumatoid arthritis, Sjgrens
syndrome, Waldenstroms disease and
pre-transplantation desensitization.
Ipilimumab is an immunostimulatory
mAb that targets cytotoxic T lymphocyte
antigen 4 (CTlA4). The candidate was
derived from Medarexs UltiMab transgenic
mouse technology and is under clinical
A marketing application was submitted
to the FDA and the European Medicines
Agency in June 2010 for the use of ipilimu-
mab as a second-line treatment for metastatic
melanoma. recent Phase III study results
indicate that ipilimumab alone or in
combination with a gp100 peptide vaccine
improved the overall survival of patients
with metastatic melanoma who had received
previous treatment
30
. Ipilimumab has been
evaluated in Phase II studies of non-small
cell lung cancer, breast cancer and prostate
cancer, as well as brain metastases. Bristol-
Myers Squibb is planning Phase III studies
of ipilimumab in non-small cell lung cancer
and in prostate cancer.
Approval success rates
Probabilities of success (POS) values, such
as cumulative approval in the United States
and transition rates between clinical phases,
have inherent limitations when the calcula-
tions involve cohorts with high percentages
of candidates in clinical study, as is the case
for human mAbs. Calculated values will vary
until fates for all candidates are known.
POS values for human mAbs that entered
clinical study after 1996 are preliminary
estimates because fates of only 31% of the
131 candidates are known (7 approved,
33 terminated), and only 20 reached
Phase III trials. In addition, the values are
likely to be underestimates because clinical
development of therapeutic mAbs takes an
average of approximately 6 years, and so the
human mAb candidates that entered clinical
study during the past 6 years have not had
sufficient time for approval.
Nevertheless, POS values are crucial for
the decision-making process used by inves-
tors, as well as for strategic planning by
the biopharmaceutical industry, and even
preliminary estimates can be useful. Based
on the current data, the cumulative approval
rate for human mAbs is 17.5%, which will
increase to 23% if raxibacumab, belimumab
and ipilimumab are approved. Transition
rates between clinical phases (which include
data for candidates currently in studies) for
the human mAbs were 89% for transitions
between Phase I to II; 51% for transitions
between Phase II to III; and 73% for transi-
tions between Phase III to approval by the
FDA (FIG. 3).
The cumulative approval success rate
for human mAbs is slightly higher than
the 15% value that we have calculated for
humanized mAbs, which first entered com-
mercial clinical development in 1988 (BOX 2).
As economic conditions, regulatory climate
and competitive landscape can change over
time, we also compared POS values for
human and humanized mAbs that entered
clinical development in the same period
(19972008). The cumulative approval
success rate was 17.5% for the human mAb
cohort (7 approvals per 40 candidates with
known fates) and 9% for the humanized
mAb cohort (5 approvals per 53 candidates
with known fates). Transition rates for
Phase I to II and Phase II to III were higher
for the human mAbs, but the Phase III
to approval rate was lower (FIG. 3). Final
fates were known for 31% and 40% of the
human and humanized mAbs developed in
this period, respectively, and the rates may
change as the final fates for more candidates
are determined.
Clinical indications
Primary therapeutic indications were
identified for the 131 human mAbs that
entered clinical study after 1996. Overall, most
mAb therapeutics, regardless of sequence
source, are developed as treatments for cancer
or immunological disorders
13
. This is also the
case for human mAbs, with 59 (45%) studied
for cancer and 36 (28%) for immunological
disorders. These proportions have remained
fairly constant since 1997 (FIG. 2).
Of the 59 antineoplastic mAbs, fates are
known for only 13 (22%): 2 are approved
products (panitumumab and ofatumumab)
and 11 candidates were terminated. The
cumulative approval success rate is 15%
based on currently available data. Most of
the 46 human antineoplastic mAbs that are
in clinical studies are at the early stages of
the process, with 5 in Phase III trials and 1
(ipilimumab) in regulatory review in the
United States and in the European Union.
Immunomodulatory human mAbs have
a higher cumulative approval success rate
(33%) than either the antineoplastic or the
Figure 2 | cumulative number of human mAbs entering clinical study between 1985 and 2008.
The primary therapeutic category for the development of human monoclonal antibodies (mAbs) that
entered clinical study sponsored by commercial firms between 1985 and 2008 was determined.
The cumulative numbers of human mAbs that entered development for antineoplastic, immuno-
modulatory, anti-infective and all other indications were tabulated. These data demonstrate the
rapid growth in human mAbs in clinical research generally, and the particularly high rates of develop-
ment of antineo plastic and immunomodulatory human mAbs.
PersPecti ves
nrd_3229_oct10.indd 770 17/9/10 11:50:13
overall human mAb cohort. This calculation
was based on data for only 12 molecules for
which definite fates are known, including
adalimumab, golimumab, canakinumab,
ustekinumab and 8 terminated candidates.
Of the 24 human immunomodulatory mAb
candidates currently in clinical development,
8 are in Phase I, 13 are in Phase II, 2 are in
Phase III and belimumab is undergoing
regulatory review in the United States and
in the European Union.
An additional 36 human mAbs that
entered clinical study after 1996 were
studied for non-traditional indications
that is, disorders that are not cancer or
immunological in nature. Half of these were
treatments for infectious diseases, with a
focus on nosocomial, anthrax and chronic
viral infections
31
. The remaining 18 candi-
dates were studied for conditions such as
osteoporosis, respiratory disorders, muscular
dystrophy, Alzheimers disease and pain.
Of the 36 candidates, 1 (denosumab) has
been approved, 1 (raxibacumab) is under-
going review by the FDA (TABLE 1) and 14
were terminated. The current cumulative
success rate for mAbs studied as treatments
for these non-traditional indications is 6.6%,
which will rise to 12.5% if raxibacumab is
approved for inhalation anthrax. The 20
candidates in clinical development were all
in either Phase I or II studies.
Molecular targets
The target of a therapeutic antibody is a
major determinant of its efficacy and safety
profile. Antigenic targets were identified
for 125 of the 131 human mAbs (TABLE 2).
These mAbs targeted a total of 89 unique
antigens, and only 22 antigens were targeted
by two or more human mAbs. Seven antigens
CTlA4, EGFr, fibronectin, insulin-like
growth factor 1 receptor (IGF1r), trans-
forming growth factor- (TGF), TNF and
TNF-related apoptosis-inducing ligand
receptor 2 (TrAIlr2) were targeted by
more than two human mAbs. Of the 125
mAbs, 55 (44%) targeted antigens that are
relevant to antineoplastic diseases, 36 (29%)
targeted antigens relevant to immunological
diseases and 17 (14%) targeted antigens
relevant to infectious diseases.
Of the 55 anticancer candidates with
known targets, 19 mAbs (33%) were
specific for only 5 targets: IGF1r (6 mAbs),
TrAIlr2 (4 mAbs), EGFr (3 mAbs),
CTlA4 (3 mAbs) and fibronectin (3 mAbs).
Of these, only EGFr was among the ten
most frequently targeted antigens for all
anticancer mAbs studied in the clinic from
1980 to 2005 (REF. 32), suggesting that devel-
opers of human mAbs may be focusing on
novel therapeutic strategies. Nevertheless,
at least some human mAbs share oncology
targets with therapeutic mAbs approved by
the FDA, including EGFr (target of panitu-
mumab and cetuximab) and CD20 (target of
rituximab, ibritumomab tiuxetan (Zevalin;
Spectrum Pharmaceuticals), iodine-131
tositumomab (Bexxar; GlaxoSmithKline)
and ofatumumab). Most of the anti-
neoplastic human mAbs target cell-surface
molecules; only two were known to target
soluble factors (hepatocyte growth factor
and platelet-derived growth factor).
By contrast, 24 out of the 36 immuno-
modulatory mAbs with known targets were
raised against cytokines and serum factors.
Interleukins constitute the largest target
group; targeted antigens include Il-1, Il-6,
Il-8, Il-12, Il-13, Il-15, Il-18, Il-17A and
Il-20. Ten out of the 36 mAbs target markers
of leukocyte activity and differentiation,
and are in early-phase clinical trials. Two
immunomodulatory human mAbs with
targets that are unique compared with those
of marketed mAbs are currently in Phase II
studies: briakinumab (developed by Abbott),
which targets the p40 subunit common to
Il-12 and Il-23, and AIN457 (developed by
Novartis), which is an Il-17A-specific mAb.
The 18 human mAbs intended for the
treatment of infectious diseases are a highly
heterogeneous group. Ten are directed
against targets implicated in viral infections,
including HIV (six mAbs), viral hepatitis
(three mAbs) and rabies (one mAb); several
of these candidates are cocktails of more
than one human mAb. Six human mAbs
target bacterial antigens, of which four are
bacterial cytotoxins, such as Clostridium
difficile enterotoxins (MK-3415A; developed
by Merck and Co.) and anthrax protective
antigen (raxibacumab), and three target
cellular features of Pseudomonas aeruginosa
Table 1 | Human mAbs approved or under FDA review*
Human mAb
(trade name;
company name)
Description indication of first
us approval
FDA
designations
Date of first us
(eu) approval
Adalimumab
(Humira; Abbott)
TNF-specific,
IgG1
rheumatoid arthritis S 31 Dec 2002
(8 Sep 2003)
Panitumumab
(vectibix; Amgen)
eGFr-specific,
IgG2
colorectal cancer P, FT, AA 27 Sep 2006
(3 Dec 2007)
Golimumab
(Simponi;
centocor)
TNF-specific,
IgG1
rheumatoid
arthritis, psoriatic
arthritis, ankylosing
spondylitis
S 24 Apr 2009
(1 Oct 2009)
canakinumab
(Ilaris; Novartis)
IL-1-specific,
IgG1
cryopyrin-
associated periodic
syndromes
P, O 18 Jun 2009
(23 Oct 2009)
Ustekinumab
(Stelara; Johnson
& Johnson)
IL-12/IL-23
p40-specific,
IgG1
Plaque psoriasis S 25 Sep 2009
(16 Jan 2009)
Ofatumumab
(Arzerra;
Genmab)
cD20-specific,
IgG1
chronic lymphocytic
leukaemia
P, FT 26 Oct 2009
(19 Apr 2010)
Denosumab
(Prolia; Amgen)
rANKL-
specific, IgG2
Treatment of
postmenopausal
osteoporosis

S 1 Jun 2010
(26 May 2010)
raxibacumab PA-specific,
IgG1
Inhalation anthrax P, FT, O Under review by
the FDA
Belimumab B lymphocyte
stimulator-
specific, IgG1
Systemic lupus
erythematosus
P, FT Under review by
the FDA and the
eMA
Ipilimumab cTLA4-specific,
IgG1
Metastatic
melanoma
P, FT, O Under review by
the FDA and the
eMA
AA, accelerated approval; cTLA, cytotoxic T lymphocyte-associated antigen; eGFr, epidermal growth
factor receptor; eMA, european Medicines Agency; eU, european Union; FDA, US Food and Drug
Administration; FT, FDA fast track drug; Ig, immunoglobulin; IL, interleukin; mAb, monoclonal antibody;
O, FDA orphan drug; P, priority review; PA, Bacillus anthracis protective antigen; rANKL, receptor for
activation of nuclear factor-B ligand; S, standard review; TNF, tumour necrosis factor. *As of June 2010.
Also approved in europe for the treatment of bone loss in patients with prostate cancer undergoing
hormone ablation therapy.
PersPecti ves
nrd_3229_oct10.indd 771 17/9/10 11:50:14
N
u
m
b
e
r

o
f

c
l
i
n
i
c
a
l

c
a
n
d
i
d
a
t
e
s
150
90
100
110
120
130
140
80
70
60
50
40
30
20
10
0
1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007
All human mAbs
Antineoplastic only
Immunomodulatory only
Anti-infective only
Other indications
sponsoring a series of Phase II studies of
belimumab in rheumatoid arthritis, Sjgrens
syndrome, Waldenstroms disease and
pre-transplantation desensitization.
Ipilimumab is an immunostimulatory
mAb that targets cytotoxic T lymphocyte
antigen 4 (CTlA4). The candidate was
derived from Medarexs UltiMab transgenic
mouse technology and is under clinical
A marketing application was submitted
to the FDA and the European Medicines
Agency in June 2010 for the use of ipilimu-
mab as a second-line treatment for metastatic
melanoma. recent Phase III study results
indicate that ipilimumab alone or in
combination with a gp100 peptide vaccine
improved the overall survival of patients
with metastatic melanoma who had received
previous treatment
30
. Ipilimumab has been
evaluated in Phase II studies of non-small
cell lung cancer, breast cancer and prostate
cancer, as well as brain metastases. Bristol-
Myers Squibb is planning Phase III studies
of ipilimumab in non-small cell lung cancer
and in prostate cancer.
Approval success rates
Probabilities of success (POS) values, such
as cumulative approval in the United States
and transition rates between clinical phases,
have inherent limitations when the calcula-
tions involve cohorts with high percentages
of candidates in clinical study, as is the case
for human mAbs. Calculated values will vary
until fates for all candidates are known.
POS values for human mAbs that entered
clinical study after 1996 are preliminary
estimates because fates of only 31% of the
131 candidates are known (7 approved,
33 terminated), and only 20 reached
Phase III trials. In addition, the values are
likely to be underestimates because clinical
development of therapeutic mAbs takes an
average of approximately 6 years, and so the
human mAb candidates that entered clinical
study during the past 6 years have not had
sufficient time for approval.
Nevertheless, POS values are crucial for
the decision-making process used by inves-
tors, as well as for strategic planning by
the biopharmaceutical industry, and even
preliminary estimates can be useful. Based
on the current data, the cumulative approval
rate for human mAbs is 17.5%, which will
increase to 23% if raxibacumab, belimumab
and ipilimumab are approved. Transition
rates between clinical phases (which include
data for candidates currently in studies) for
the human mAbs were 89% for transitions
between Phase I to II; 51% for transitions
between Phase II to III; and 73% for transi-
tions between Phase III to approval by the
FDA (FIG. 3).
The cumulative approval success rate
for human mAbs is slightly higher than
the 15% value that we have calculated for
humanized mAbs, which first entered com-
mercial clinical development in 1988 (BOX 2).
As economic conditions, regulatory climate
and competitive landscape can change over
time, we also compared POS values for
human and humanized mAbs that entered
clinical development in the same period
(19972008). The cumulative approval
success rate was 17.5% for the human mAb
cohort (7 approvals per 40 candidates with
known fates) and 9% for the humanized
mAb cohort (5 approvals per 53 candidates
with known fates). Transition rates for
Phase I to II and Phase II to III were higher
for the human mAbs, but the Phase III
to approval rate was lower (FIG. 3). Final
fates were known for 31% and 40% of the
human and humanized mAbs developed in
this period, respectively, and the rates may
change as the final fates for more candidates
are determined.
Clinical indications
Primary therapeutic indications were
identified for the 131 human mAbs that
entered clinical study after 1996. Overall, most
mAb therapeutics, regardless of sequence
source, are developed as treatments for cancer
or immunological disorders
13
. This is also the
case for human mAbs, with 59 (45%) studied
for cancer and 36 (28%) for immunological
disorders. These proportions have remained
fairly constant since 1997 (FIG. 2).
Of the 59 antineoplastic mAbs, fates are
known for only 13 (22%): 2 are approved
products (panitumumab and ofatumumab)
and 11 candidates were terminated. The
cumulative approval success rate is 15%
based on currently available data. Most of
the 46 human antineoplastic mAbs that are
in clinical studies are at the early stages of
the process, with 5 in Phase III trials and 1
(ipilimumab) in regulatory review in the
United States and in the European Union.
Immunomodulatory human mAbs have
a higher cumulative approval success rate
(33%) than either the antineoplastic or the
Figure 2 | cumulative number of human mAbs entering clinical study between 1985 and 2008.
The primary therapeutic category for the development of human monoclonal antibodies (mAbs) that
entered clinical study sponsored by commercial firms between 1985 and 2008 was determined.
The cumulative numbers of human mAbs that entered development for antineoplastic, immuno-
modulatory, anti-infective and all other indications were tabulated. These data demonstrate the
rapid growth in human mAbs in clinical research generally, and the particularly high rates of develop-
ment of antineo plastic and immunomodulatory human mAbs.
PersPecti ves
nrd_3229_oct10.indd 770 17/9/10 11:50:13
based on results for a small number of
candidates. This rate is lower than that
observed for all human mAbs (17.5%) and for
the transgenic-mouse-derived human mAbs
(29%); however, if raxibacumab and belimu-
mab are approved, the cumulative approval
success rate will increase to 30%. Human
mAbs derived from phage-display technol-
ogy have high Phase I to II and Phase II to III
transition rates, but the Phase III to approval
rate is currently 50% (FIG. 3).
None of the 35 mAbs known to be
derived from phage display was identified
as IgG2, which is an interesting observa-
tion given the diversity of phage-display
platforms. Of those with known isotypes,
20 were IgG1 and 9 were IgG4. Cambridge
Antibody Technology was responsible for
15 phage-display-derived human mAbs,
including adalimumab and four terminated
candidates; these therefore have a cumula-
tive success rate of 20% so far, which is
comparable to that of Medarex platform-
derived mAbs. Cambridge Antibody
Technology has also produced a substantial
number of IgG4 molecules (7 out of 9
IgG4 mAbs are known to be derived from
phage-display technology).
Conclusions
The acquisition of mAb technology com-
panies, including Abgenix, Cambridge
Antibody Technology and Medarex, by
major drug companies is an indication of
the pharmaceutical industrys increasing
interest in human mAb therapeutics.
Our analyses indicate that human mAbs
are a rich source of new therapeutics, with
7 approved in the United States, 3 under
review by the FDA, 7 in late-stage develop-
ment and 81 in early-stage development.
Although limited data are available, the cur-
rent POS rates for human mAbs are similar
or superior to those for current humanized
mAb candidates. The cumulative approval
success rate for human mAbs is currently
17.5%, although this could change substan-
tially in the coming years depending on the
fate of the high percentage of candidates
that are still in clinical study.
We found that human mAbs are primarily
in development for the treatment of cancer
and immunological disorders. The cumula-
tive approval rate for immunomodulatory
human mAbs (33%) is higher than that for
the antineoplastic (15%) candidates. This
difference is also observed when chimeric
and humanized therapeutic mAbs are con-
sidered
1
. The majority of human mAbs were
derived from transgenic mouse technolo-
gies and from phage-display technologies,
although human hybridoma and
transformed cells have also been used
to produce human mAbs.
The two transgenic mouse technologies
have so far generated six mAbs that gained
approval (29% POS) and one candidate that
is under review by the FDA. The hetero-
geneous mix of phage-display technologies
collectively have generated one marketed
product (12.5% POS) and two candidates
that are under review by the FDA. The
earliest transgenic-mouse-derived and
phage-display-derived human mAbs entered
clinical development in the same year, so
timing cannot account for the differences.
However, if raxibacumab, belimumab and
ipilimumab are approved, phage-display-
derived human mAbs will have demonstrated
preliminary POS rates comparable to
mouse-derived human mAbs (30% versus
32%, respectively). A potential disincentive
to the use of the transgenic mouse platform
is that the intellectual property is controlled
by only a few companies, and so access may
be costly. Phage-display technologies are
particularly advantageous when target
antigens are shared by humans and mice.
With the current trend towards devel-
oping targeted therapeutics, the focus on
human mAbs is likely to intensify owing
to a perceived low level of immunogenicity
of these agents (BOX 1). The pharmaceutical
and biotechnology industry, regulatory
agencies, physicians and patients have now
gained sufficient experience with mAbs
to view them as little different from any
other therapeutic. The data so far indicate
that mAbs derived from human sequences
by various technologies are effective in
addressing novel therapeutic targets, and
Table 2 | Antigenic targets of human mAbs in development*
Antigen Therapeutic category No. of human mAbs
Bacillus anthracis PA Infectious disease 2
cD30 cancer 2
cD40 cancer 2
Clostridium difficile enterotoxin Infectious disease 2
cTLA4 cancer 3
eGFr cancer 3
ePcAM cancer 2
Fibronectin cancer 3
GM-cSF Immunological disease 2
Her3 cancer 2
HIv gp41 Infectious disease 2
IGF1r cancer 6
IL-6 Immunological disease 2
Integrins Immunological disease 2
PDGF cancer, immunological disease 2
PSMA cancer 2
Tenascin cancer 2
TGF
Immunological, ophthalmic
and fibrotic diseases
3
TNF Immunological disease 3
TrAILr2 cancer 4
Unknown cancer, immunological disease 6
cTLA4, cytotoxic T lymphocyte-associated antigen 4; eGFr, epidermal growth factor receptor;
ePcAM, epithelial cell adhesion molecule; GM-cSF, granulocytemacrophage colony-stimulating factor;
gp, glycoprotein; Her3, human epidermal growth factor receptor 3 (also known as erBB3); IGF1r,
insulin-like growth factor 1 receptor; IL, interleukin; mAb, monoclonal antibody; PA, protective antigen;
PDGF, platelet-derived growth factor; PSMA, prostate-specific membrane antigen (also known as FOLH1);
TGF, transforming growth factor; TNF, tumour necrosis factor; TrAILr2, tumour necrosis factor-related
apoptosis-inducing ligand receptor 2 (also known as TNFrSF10B). *This table lists molecules that were
targets for a minimum of two human mAbs that entered clinical study between 1997 and 2008.
PersPecti ves
nrd_3229_oct10.indd 773 17/9/10 11:50:15
T
r
a
n
s
i
t
i
o
n

r
a
t
e

(
%
)
100
90
80
70
60
50
40
30
20
10
0
Phase III
US approval
Phase IIIII Phase III
80
89
87
94
47
51
60
73
86
73
86
50
Humanized, 19972008 (n = 133)
Human, 19972008 (n = 131)
Transgenic mouse (n = 56)
Display technology (n = 35)
(KB001; developed by Kalobios/Sanofi
Pasteur, and panobacumab; developed by
Kenta Biotech) and methicillin-resistant
Staphylococcus aureus (Aurograb; devel-
oped by Novartis). Finally, one human mAb
(efungumab; developed by Novartis) targets
candidal heat shock protein 90, an intra-
cellular antigen released during infection.
Platform technologies
We were able to identify the platform
technologies that were used to develop
103 (79%) of the 131 mAbs analysed. Despite
the availability of innovative approaches to
sample natural human immune responses
for the creation of mAbs
10
, most therapeutic
human mAbs in clinical study were derived
from either immunization of transgenic
mice expressing human antibody genes
or phage-display recombinants. The first
candidate molecules from both technolo-
gies entered clinical development in the late
1990s, so performance differences cannot
be attributed to time-dependent variables.
Use of transgenic mice expressing
human immunoglobulins avoids human
anti-mouse antibody responses and main-
tains the technical advantages of mouse
hybridomas. Of the 103 candidates from
identified platforms, 56 were produced
in transgenic mice; 6 were approved for
marketing (panitumumab, golimumab,
canakinumab, ustekinumab, ofatumumab
and denosumab), and 15 were terminated.
The current cumulative approval success
rate is therefore 29%, a higher rate than that
currently calculated for human mAbs as
a whole (17.5%). In particular, candidates
derived from transgenic mouse platforms
have considerably higher Phase II to III and
Phase III to approval transition rates than
those of the entire cohort of human mAbs
(FIG. 3). Again, it is important to note that
these rates will change to some extent over
time as fates for more human mAb candi-
dates are determined in the future. The two
primary technologies for generating human
mAbs from transgenic mice were first
described in 1994 (REFs 33,34) and use
two different engineering approaches to
inactivate endogenous mouse genes and
to insert exogenous human immunoglobulin
genes
35,36
.
A total of 34 human mAbs were identified
as being derived from Medarexs HuMAb,
UltiMab, TC Mouse or KM Mouse platforms.
Four of these candidates have been approved
(golimumab, canakinumab, ustekinumab
and ofatumumab), 1 (ipilimumab) is under
regulatory review, 7 were terminated and
22 are now in clinical studies (7 at Phase I,
14 at Phase II and 1 at Phase III). Although
immunoglobulin isotype was not known for
all the candidates, the majority of Medarex
platform-derived molecules were IgG1
(at least 21 mAbs). Of the total 34 mAbs,
16 (47%) were intended for the treatment
of cancer, 13 (38%) for immunological
conditions and 3 (9%) were anti-infective
agents.
The XenoMouse platform developed by
Abgenix, which was acquired by Amgen
in 2005, was used for the development of
at least 18 human mAbs. Two (panitumu-
mab and denosumab) are approved, nine
are in clinical study (two at Phase II, five
at Phase II and two at Phase III) and seven
are discontinued. Most of the XenoMouse-
derived candidates were IgG2 (11 mAbs),
four were IgG1, one was IgG4 and two were
of unknown isotype. Two-thirds of the mAbs
were cancer treatments; only two (11%) were
for immunological conditions and one (6%)
was an anti-infective agent.
Other transgenic mouse-based platforms,
such as regenerons VelocImmune and XTl
Biopharmaceuticals Trimera, have each
yielded at least one early-stage candidate.
A second popular approach for producing
human mAbs is the recombinant expression
of human antigen-binding fragments in a
bacteriophage and subsequent selection is
based on desirable antigen-binding proper-
ties
3739
. This technology was used to create
at least 35 human mAbs that entered clinical
development. Unlike transgenic mouse
technologies, phage-display is used
by numerous companies, including
MedImmune Cambridge (formerly
Cambridge Antibody Technology), Dyax,
MorphoSys, BioInvent and NeuTec.
One phage-display-derived mAb
(adalimumab) has been approved, and two
(raxibacumab and belimumab) are under
review by the FDA. In addition, 3 phage-
display-derived mAb candidates are in
Phase I, 19 are in Phase II and 3 are in Phase
III. So far, development of seven mAbs has
been discontinued. The current cumulative
approval success rate for all phage-display-
based technologies is 12.5%, although this is
Figure 3 | Transition rates between clinical
phases for human mAbs. The historical rates
of transition from Phase I to II, Phase II to III and
Phase III to review by the US Food and Drug
Administration are depicted. The review to
approval rate was 100% for all human mono-
clonal antibodies (mAbs) in the categories
presented. Data for human mAbs derived from
transgenic mouse and display technologies
are shown separately and data for humanized
mAbs are included for comparison.
glossary
Allotype
Antibody allotypes are defined by their polymorphism
within the immunoglobulin heavy and light chains. Natural
allelic genetic variation in the constant region of genes
in humans may predispose a given patient to anti-drug
antibody responses if the drug is a foreign allotype.
Ankylosing spondylitis
A chronic condition of unknown aetiology that is
characterized by inflammation of the joints of the spine
and pelvis. Disease progression may result in fusion of
the joints.
Anti-idiotypic antibody
An antibody that targets the hypervariable antigen-binding
domain of an exogenous immunoglobulin, including
therapeutic monoclonal antibodies. As the constant regions
are fairly conserved, with the exception of allotypic
differences, many anti-immunoglobulin responses will be
directed against the highly variable, antigen-binding domain.
Cryopyrin-associated periodic syndromes
(CAPs). A group of rare, inherited autoimmune disorders
associated with over-secretion of interleukin-1 that may
cause inflammation of the skin, eyes, bones, joints and
meninges.
Phage-display technologies
A method involving the use of bacteriophages to select
desirable antibody variable domains based on their
binding properties.
Pre-transplant desensitization
In the recipient patient, reduction of antibody-
producing cells or the amount of circulating antibodies
that might target foreign tissue prior to transplantation
of an organ.
Systemic lupus erythematosus
A chronic, inflammatory autoimmune disease affecting
connective tissue throughout the body.
glossary
Glossary head
Glossary body
Glossary head
Glossary body
Glossary head
Glossary body
Glossary head
Glossary body
Glossary head
Glossary body
PersPecti ves
nrd_3229_oct10.indd 772 17/9/10 11:50:14
based on results for a small number of
candidates. This rate is lower than that
observed for all human mAbs (17.5%) and for
the transgenic-mouse-derived human mAbs
(29%); however, if raxibacumab and belimu-
mab are approved, the cumulative approval
success rate will increase to 30%. Human
mAbs derived from phage-display technol-
ogy have high Phase I to II and Phase II to III
transition rates, but the Phase III to approval
rate is currently 50% (FIG. 3).
None of the 35 mAbs known to be
derived from phage display was identified
as IgG2, which is an interesting observa-
tion given the diversity of phage-display
platforms. Of those with known isotypes,
20 were IgG1 and 9 were IgG4. Cambridge
Antibody Technology was responsible for
15 phage-display-derived human mAbs,
including adalimumab and four terminated
candidates; these therefore have a cumula-
tive success rate of 20% so far, which is
comparable to that of Medarex platform-
derived mAbs. Cambridge Antibody
Technology has also produced a substantial
number of IgG4 molecules (7 out of 9
IgG4 mAbs are known to be derived from
phage-display technology).
Conclusions
The acquisition of mAb technology com-
panies, including Abgenix, Cambridge
Antibody Technology and Medarex, by
major drug companies is an indication of
the pharmaceutical industrys increasing
interest in human mAb therapeutics.
Our analyses indicate that human mAbs
are a rich source of new therapeutics, with
7 approved in the United States, 3 under
review by the FDA, 7 in late-stage develop-
ment and 81 in early-stage development.
Although limited data are available, the cur-
rent POS rates for human mAbs are similar
or superior to those for current humanized
mAb candidates. The cumulative approval
success rate for human mAbs is currently
17.5%, although this could change substan-
tially in the coming years depending on the
fate of the high percentage of candidates
that are still in clinical study.
We found that human mAbs are primarily
in development for the treatment of cancer
and immunological disorders. The cumula-
tive approval rate for immunomodulatory
human mAbs (33%) is higher than that for
the antineoplastic (15%) candidates. This
difference is also observed when chimeric
and humanized therapeutic mAbs are con-
sidered
1
. The majority of human mAbs were
derived from transgenic mouse technolo-
gies and from phage-display technologies,
although human hybridoma and
transformed cells have also been used
to produce human mAbs.
The two transgenic mouse technologies
have so far generated six mAbs that gained
approval (29% POS) and one candidate that
is under review by the FDA. The hetero-
geneous mix of phage-display technologies
collectively have generated one marketed
product (12.5% POS) and two candidates
that are under review by the FDA. The
earliest transgenic-mouse-derived and
phage-display-derived human mAbs entered
clinical development in the same year, so
timing cannot account for the differences.
However, if raxibacumab, belimumab and
ipilimumab are approved, phage-display-
derived human mAbs will have demonstrated
preliminary POS rates comparable to
mouse-derived human mAbs (30% versus
32%, respectively). A potential disincentive
to the use of the transgenic mouse platform
is that the intellectual property is controlled
by only a few companies, and so access may
be costly. Phage-display technologies are
particularly advantageous when target
antigens are shared by humans and mice.
With the current trend towards devel-
oping targeted therapeutics, the focus on
human mAbs is likely to intensify owing
to a perceived low level of immunogenicity
of these agents (BOX 1). The pharmaceutical
and biotechnology industry, regulatory
agencies, physicians and patients have now
gained sufficient experience with mAbs
to view them as little different from any
other therapeutic. The data so far indicate
that mAbs derived from human sequences
by various technologies are effective in
addressing novel therapeutic targets, and
Table 2 | Antigenic targets of human mAbs in development*
Antigen Therapeutic category No. of human mAbs
Bacillus anthracis PA Infectious disease 2
cD30 cancer 2
cD40 cancer 2
Clostridium difficile enterotoxin Infectious disease 2
cTLA4 cancer 3
eGFr cancer 3
ePcAM cancer 2
Fibronectin cancer 3
GM-cSF Immunological disease 2
Her3 cancer 2
HIv gp41 Infectious disease 2
IGF1r cancer 6
Integrins Immunological disease 2
PDGF cancer, immunological disease 2
PSMA cancer 2
Tenascin cancer 2
TGF
Immunological, ophthalmic
and fibrotic diseases
3
TNF Immunological disease 3
TrAILr2 cancer 4
Unknown cancer, immunological disease 6
cTLA4, cytotoxic T lymphocyte-associated antigen 4; eGFr, epidermal growth factor receptor;
ePcAM, epithelial cell adhesion molecule; GM-cSF, granulocytemacrophage colony-stimulating factor;
gp, glycoprotein; Her3, human epidermal growth factor receptor 3 (also known as erBB3); IGF1r,
insulin-like growth factor 1 receptor; IL, interleukin; mAb, monoclonal antibody; PA, protective antigen;
PDGF, platelet-derived growth factor; PSMA, prostate-specific membrane antigen (also known as FOLH1);
TGF, transforming growth factor; TNF, tumour necrosis factor; TrAILr2, tumour necrosis factor-related
apoptosis-inducing ligand receptor 2 (also known as TNFrSF10B). *This table lists molecules that were
targets for a minimum of two human mAbs that entered clinical study between 1997 and 2008.
PersPecti ves
nrd_3229_oct10.indd 773 17/9/10 11:50:15
T
r
a
n
s
i
t
i
o
n

r
a
t
e

(
%
)
100
90
80
70
60
50
40
30
20
10
0
Phase III
US approval
Phase IIIII Phase III
80
89
87
94
47
51
60
73
86
73
86
50
Humanized, 19972008 (n = 133)
Human, 19972008 (n = 131)
Transgenic mouse (n = 56)
Display technology (n = 35)
(KB001; developed by Kalobios/Sanofi
Pasteur, and panobacumab; developed by
Kenta Biotech) and methicillin-resistant
Staphylococcus aureus (Aurograb; devel-
oped by Novartis). Finally, one human mAb
(efungumab; developed by Novartis) targets
candidal heat shock protein 90, an intra-
cellular antigen released during infection.
Platform technologies
We were able to identify the platform
technologies that were used to develop
103 (79%) of the 131 mAbs analysed. Despite
the availability of innovative approaches to
sample natural human immune responses
for the creation of mAbs
10
, most therapeutic
human mAbs in clinical study were derived
from either immunization of transgenic
mice expressing human antibody genes
or phage-display recombinants. The first
candidate molecules from both technolo-
gies entered clinical development in the late
1990s, so performance differences cannot
be attributed to time-dependent variables.
Use of transgenic mice expressing
human immunoglobulins avoids human
anti-mouse antibody responses and main-
tains the technical advantages of mouse
hybridomas. Of the 103 candidates from
identified platforms, 56 were produced
in transgenic mice; 6 were approved for
marketing (panitumumab, golimumab,
canakinumab, ustekinumab, ofatumumab
and denosumab), and 15 were terminated.
The current cumulative approval success
rate is therefore 29%, a higher rate than that
currently calculated for human mAbs as
a whole (17.5%). In particular, candidates
derived from transgenic mouse platforms
have considerably higher Phase II to III and
Phase III to approval transition rates than
those of the entire cohort of human mAbs
(FIG. 3). Again, it is important to note that
these rates will change to some extent over
time as fates for more human mAb candi-
dates are determined in the future. The two
primary technologies for generating human
mAbs from transgenic mice were first
described in 1994 (REFs 33,34) and use
two different engineering approaches to
inactivate endogenous mouse genes and
to insert exogenous human immunoglobulin
genes
35,36
.
A total of 34 human mAbs were identified
as being derived from Medarexs HuMAb,
UltiMab, TC Mouse or KM Mouse platforms.
Four of these candidates have been approved
(golimumab, canakinumab, ustekinumab
and ofatumumab), 1 (ipilimumab) is under
regulatory review, 7 were terminated and
22 are now in clinical studies (7 at Phase I,
14 at Phase II and 1 at Phase III). Although
immunoglobulin isotype was not known for
all the candidates, the majority of Medarex
platform-derived molecules were IgG1
(at least 21 mAbs). Of the total 34 mAbs,
16 (47%) were intended for the treatment
of cancer, 13 (38%) for immunological
conditions and 3 (9%) were anti-infective
agents.
The XenoMouse platform developed by
Abgenix, which was acquired by Amgen
in 2005, was used for the development of
at least 18 human mAbs. Two (panitumu-
mab and denosumab) are approved, nine
are in clinical study (two at Phase II, five
at Phase II and two at Phase III) and seven
are discontinued. Most of the XenoMouse-
derived candidates were IgG2 (11 mAbs),
four were IgG1, one was IgG4 and two were
of unknown isotype. Two-thirds of the mAbs
were cancer treatments; only two (11%) were
for immunological conditions and one (6%)
was an anti-infective agent.
Other transgenic mouse-based platforms,
such as regenerons VelocImmune and XTl
Biopharmaceuticals Trimera, have each
yielded at least one early-stage candidate.
A second popular approach for producing
human mAbs is the recombinant expression
of human antigen-binding fragments in a
bacteriophage and subsequent selection is
based on desirable antigen-binding proper-
ties
3739
. This technology was used to create
at least 35 human mAbs that entered clinical
development. Unlike transgenic mouse
technologies, phage-display is used
by numerous companies, including
MedImmune Cambridge (formerly
Cambridge Antibody Technology), Dyax,
MorphoSys, BioInvent and NeuTec.
One phage-display-derived mAb
(adalimumab) has been approved, and two
(raxibacumab and belimumab) are under
review by the FDA. In addition, 3 phage-
display-derived mAb candidates are in
Phase I, 19 are in Phase II and 3 are in Phase
III. So far, development of seven mAbs has
been discontinued. The current cumulative
approval success rate for all phage-display-
based technologies is 12.5%, although this is
Figure 3 | Transition rates between clinical
phases for human mAbs. The historical rates
of transition from Phase I to II, Phase II to III and
Phase III to review by the US Food and Drug
Administration are depicted. The review to
approval rate was 100% for all human mono-
clonal antibodies (mAbs) in the categories
presented. Data for human mAbs derived from
transgenic mouse and display technologies
are shown separately and data for humanized
mAbs are included for comparison.
glossary
Allotype
Antibody allotypes are defined by their polymorphism
within the immunoglobulin heavy and light chains. Natural
allelic genetic variation in the constant region of genes
in humans may predispose a given patient to anti-drug
antibody responses if the drug is a foreign allotype.
Ankylosing spondylitis
A chronic condition of unknown aetiology that is
characterized by inflammation of the joints of the spine
and pelvis. Disease progression may result in fusion of
the joints.
Anti-idiotypic antibody
An antibody that targets the hypervariable antigen-binding
domain of an exogenous immunoglobulin, including
therapeutic monoclonal antibodies. As the constant regions
are fairly conserved, with the exception of allotypic
differences, many anti-immunoglobulin responses will be
directed against the highly variable, antigen-binding domain.
Cryopyrin-associated periodic syndromes
(CAPs). A group of rare, inherited autoimmune disorders
associated with over-secretion of interleukin-1 that may
cause inflammation of the skin, eyes, bones, joints and
meninges.
Phage-display technologies
A method involving the use of bacteriophages to select
desirable antibody variable domains based on their
binding properties.
Pre-transplant desensitization
In the recipient patient, reduction of antibody-
producing cells or the amount of circulating antibodies
that might target foreign tissue prior to transplantation
of an organ.
Systemic lupus erythematosus
A chronic, inflammatory autoimmune disease affecting
connective tissue throughout the body.
glossary
Glossary head
Glossary body
Glossary head
Glossary body
Glossary head
Glossary body
Glossary head
Glossary body
Glossary head
Glossary body
PersPecti ves
nrd_3229_oct10.indd 772 17/9/10 11:50:14
If it were not for the great variability
among individuals, medicine might
as well be a science and not an art.
Sir William Osler, 1892
Patients and society expect that the benefits
of licensed drugs will outweigh their risks.
Drug regulations in all major jurisdictions
are aiming to ensure a positive benefitrisk
balance. However, regulatory decisions are
based on population-level information, with
an understanding that the benefitrisk will
not necessarily be positive for all treated
patients. Patients are not equally responsive
to beneficial effects, and not equally sus-
ceptible to adverse effects. Drug developers,
drug regulators and health-care profession-
als have been aware of patient heterogeneity
and that one size doesnt fit all. Regulators
attempt to direct prescribers to appropri-
ate patient subpopulations and treatment
scenarios, whereas health-care professionals
aim to tailor drug regimens to their indi-
vidual patients, often based on phenotypic
markers that are expected to influence a
drugs pharmacokinetics or pharmacody-
namics such as creatinine clearance, disease
stage or comorbidities. In some cases, pre-
scribers can tailor doses to individuals by
titrating to their response to thedrug.
In spite of such efforts, drugs often do
not perform as well in clinical practice as in
the clinical trials that provide the basis for
marketing authorization. This difference
in benefitrisk has been termed the effi-
cacyeffectiveness gap, in which efficacy is
defined as the extent to which an interven-
tion does more good than harm under ideal
circumstances (that is, under clinical trial
conditions), whereas effectiveness is the
extent to which an intervention does more
good than harm when provided under the
usual circumstances of health care practice
(REF.1). Others have referred to efficacy as
can it work? and effectiveness as does it
work? (REF.2).
There have been several instances in
which the quality of prescription drugs dif-
fered markedly between the clinical trial
medicine assessed by regulators and that
being marketed. For example, some marketed
heparin products became contaminated with
oversulphated chondroitin sulphate
3
, and the
benefitrisk of nelfinavir (Viracept; Pfizer)
came under intense scrutiny when batches
of the medicine became contaminated with a
known genotoxic substance
4
. However, cases
in which pharmaceutical quality issues give
rise to differences between efficacy and effec-
tiveness are rare and represent a distinct set
of regulatory problems; therefore they are not
further discussedhere.
Another reason for an apparent difference
between efficacy and effectiveness may be the
inadequate follow up of minor safety signals
that arise during pre-licensing develop-
ment. We re-emphasise the responsibility of
manufacturers and regulators to ensure thor-
ough post-licensing investigation of safety
signals; it is hoped that the introduction
of Risk Management Plans (RMPs) in the
European Union (EU) and Risk Evaluation
and Mitigation Strategies (REMSs) in the
United States a few years ago may success-
fully address this important issue
5
. However,
in this Opinion article we argue that the
efficacyeffectiveness gap is, in most cases, a
problem of variability in drug response. We
describe biological and behavioural sources
of variability. We then summarize current
attempts to bridge the efficacyeffectiveness
gap by reducing variability. Finally, we specu-
late on new opportunities and technologies
in this area and what regulators can do to
advance the best use ofdrugs.
Variability and treatment scenarios
We observe that over the past decade there
were few, if any, drugs that came under close
scrutiny for their benefitrisk profile or
that had to be withdrawn from the market
OPI NI ON
Bridging the efficacyeffectiveness
gap: a regulators perspective
on addressing variability of drug
response
Hans-Georg Eichler, Eric Abadie, Alasdair Breckenridge, Bruno Flamion,
Lars L.Gustafsson, Hubert Leufkens, Malcolm Rowland, Christian K.Schneider
and Brigitte Bloechl-Daum
Abstract | Drug regulatory agencies should ensure that the benefits of drugs
outweigh their risks, but licensed medicines sometimes do not perform as
expected in everyday clinical practice. Failure may relate to lower than anticipated
efficacy or a higher than anticipated incidence or severity of adverse effects. Here
we show that the problem of benefitrisk is to a considerable degree a problem of
variability in drug response. We describe biological and behavioural sources of
variability and how these contribute to the long-known efficacyeffectiveness gap.
In this context, efficacy describes how a drug performs under conditions of clinical
trials, whereas effectiveness describes how it performs under conditions of
everyday clinical practice. We argue that a broad range of pre- and post-licensing
technologies will need to be harnessed to bridge the efficacyeffectiveness gap.
Successful approaches will not be limited to the current notion of pharmaco-
genomics-based personalized medicines, but will also entail the wider use of
electronic health-care tools to improve drug prescribing and patient adherence.
PERSPECTIVES
nrd_3501_jul11.indd 495 20/06/2011 16:34
are likely to be less immunogenic than those
with rodent-derived sequences. There is
considerable unmet medical need in the
three main areas of study for investigational
human mAbs cancer, immunological and
infectious diseases and these emerging
agents could therefore provide valuable new
treatment options.
Aaron L. Nelson was previously at the Tufts University
School of Medicine, Boston, Massachusetts 02118, USA.
Present address: Novartis Institutes for Biomedical
Research, Room 7226, 7th floor, 300 Technology
Square, Cambridge, Massachusetts 02139, USA.
Eugen Dhimolea is at the Tufts University
School of Medicine, 136 Harrison Avenue, Boston,
Massachusetts 02111, USA.
Janice M. Reichert is at the Tufts Center for the Study
of Drug Development, Suite 1100, 75 Kneeland Street,
Boston, Massachusetts 02111, USA.
Correspondence to J.M.R.
email: janice.reichert@tufts.edu
doi:10.1038/nrd3229
Published online 3 september 2010
1. Reichert, J. M. Monoclonal antibodies as innovative
therapeutics. Curr. Pharm. Biotechnol. 9, 423430
(2008).
2. Reichert, J. M., Rosensweig, C. J., Faden, L. B. &
Dewitz, M. C. Monoclonal antibody successes in the
clinic. Nature Biotech. 23, 10731078 (2005).
3. Reichert, J. M. Antibodies to watch in 2010. MAbs
2, 84100 (2010).
4. James, K. & Bell, G. T. Human monoclonal antibody
production. Current status and future prospects.
J. Immunol. Methods 100, 540 (1987).
5. Olsson, L. & Kaplan, H. S. Humanhuman hybridomas
producing monoclonal antibodies of predefined
antigenic specificity. Proc. Natl Acad. Sci. USA 77,
54295431 (1980).
6. Shoenfeld, Y. et al. Production of autoantibodies
by humanhuman hybridomas. J. Clin. Invest. 70,
205208 (1982).
7. Olsson, L. et al. Antibody producing humanhuman
hybridomas. II. Derivation and characterization
of an antibody specific for human leukemia cells.
J. Exp. Med. 159, 537550 (1984).
8. Kozbor, D. & Roder, J. C. Requirements for the
establishment of high-titered human monoclonal
antibodies against tetanus toxoid using the
EpsteinBarr virus technique. J. Immunol. 127,
12751280 (1981).
9. Kozbor, D., Lagarde, A. E. & Roder, J. C.
Human hybridomas constructed with antigen-specific
EpsteinBarr virus-transformed cell lines. Proc. Natl
Acad. Sci. USA 79, 66516655 (1982).
10. Beerli, R. R. & Rader, C. Mining human antibody
repertoires. MAbs 2, 361374 (2010).
11. Teng, N. N. et al. Protection against Gram-negative
bacteremia and endotoxemia with human
monoclonal IgM antibodies. Proc. Natl Acad. Sci.
USA 82, 17901794 (1985).
12. Ziegler, E. J. et al. Treatment of Gram-negative
bacteremia and septic shock with HA-1A human
monoclonal antibody against endotoxin. A
randomized, double-blind, placebo-controlled trial.
The HA-1A Sepsis Study Group. N. Engl. J. Med.
324, 429436 (1991).
13. Cross, A. S. Antiendotoxin antibodies: a dead end?
Ann. Intern. Med. 121, 5860 (1994).
14. Luce, J. M. Introduction of new technology into critical
care practice: a history of HA-1A human monoclonal
antibody against endotoxin. Crit. Care Med. 21,
12331241 (1993).
15. McCloskey, R. V., Straube, R. C., Sanders, C.,
Smith, S. M. & Smith, C. R. Treatment of septic shock
with human monoclonal antibody HA-1A. A
CHESS Trial Study Group. Ann. Intern. Med. 121, 15
(1994).
16. Abbott. Abbott Annual Report 2008. Abbott website
[online], http://www.abbott.com/static/content/microsite/
annual_report/2008/16_review1.html (2008).
17. US Food and Drug Administration. Vectibix
Panitumumab Injectable. Application No.: 125147.
Medical Review(s). FDA website [online],
http://www.accessdata.fda.gov/drugsatfda_docs/
nda/2006/125147s0000_MedR.pdf (2006).
18. Van Cutsem, E. et al. Open-label phase III trial of
panitumumab plus best supportive care compared
with best supportive care alone in patients with
chemotherapy-refractory metastatic colorectal cancer.
J. Clin. Oncol. 25, 16581664 (2007).
19. Mazumdar, S. & Greenwald, D. Golimumab. MAbs
1, 422431 (2009).
20. Lachmann, H. J. et al. Use of canakinumab in the
cryopyrin-associated periodic syndrome. N. Engl.
J. Med. 360, 24162425 (2009).
21. Dhimolea, E. Canakinumab. MAbs 2, 313 (2010).
22. Cingoz, O. Ustekinumab. MAbs 1, 216221 (2009).
23. Teeling, J. L. et al. The biological activity of human
epitopes on CD20. J. Immunol. 177, 362371 (2006).
24. Glennie, M. J., French, R. R., Cragg, M. S. & Taylor, R. P.
Mechanisms of killing by anti-CD20 monoclonal
antibodies. Mol. Immunol. 44, 38233837 (2007).
25. Zhang, B. Ofatumumab. MAbs 1, 326331 (2009).
26. Pageau, S. C. Denosumab. MAbs 1, 210215 (2009).
27. Mazumdar, S. Raxibacumab. MAbs 1, 531538
(2009).
28. DallEra, M. & Wofsy, D. Connective tissue diseases:
belimumab for systemic lupus erythematosus: breaking
through? Nature Rev. Rheumatol. 6, 124125 (2010).
29. Wallace, D. J. et al. A phase II, randomized, double-
blind, placebo-controlled, dose-ranging study of
belimumab in patients with active systemic lupus
erythematosus. Arthritis Rheum. 61, 11681178
(2009).
30. Hodi, F. S. et al. Improved survival with ipilimumab in
patients with metastatic melanoma. N. Engl. J. Med.
363, 711723 (2010).
31. Reichert, J. M. & Dewitz, M. C. Anti-infective
monoclonal antibodies: perils and promise of
development. Nature Rev. Drug Discov. 5, 191195
(2006).
32. Reichert, J. M. & Valge-Archer, V. E. Development
trends for monoclonal antibody cancer therapeutics.
33. Green, L. L. et al. Antigen-specific human monoclonal
antibodies from mice engineered with human Ig
heavy and light chain YACs. Nature Genet. 7, 1321
(1994).
34. Lonberg, N. et al. Antigen-specific human antibodies
from mice comprising four distinct genetic
modifications. Nature 368, 856859 (1994).
35. Green, L. L. Antibody engineering via genetic
engineering of the mouse: XenoMouse strains are a
vehicle for the facile generation of therapeutic human
monoclonal antibodies. J. Immunol. Methods 231,
1123 (1999).
36. Lonberg, N. Human antibodies from transgenic
animals. Nature Biotech. 23, 11171125 (2005).
37. McCafferty, J., Griffiths, A. D., Winter, G. & Chiswell,
D. J. Phage antibodies: filamentous phage displaying
antibody variable domains. Nature 348, 552554
(1990).
38. Vaughan, T. J. et al. Human antibodies with
sub-nanomolar affinities isolated from a large
non-immunized phage display library. Nature Biotech.
14, 309314 (1996).
39. Clackson, T., Hoogenboom, H. R., Griffiths, A. D. &
Winter, G. Making antibody fragments using phage
display libraries. Nature 352, 624628 (1991).
40. Chirino, A. J., Ary, M. L. & Marshall, S. A.
Minimizing the immunogenicity of protein
therapeutics. Drug Discov. Today 9, 8290 (2004).
41. Baert, F. et al. Influence of immunogenicity on the
long-term efficacy of infliximab in Crohns disease.
N. Engl. J. Med. 348, 601608 (2003).
42. De Groot, A. S. & Scott, D. W. Immunogenicity of
protein therapeutics. Trends Immunol. 28, 482490
(2007).
43. Bartelds, G. M. et al. Clinical response to
adalimumab: relationship to anti-adalimumab
antibodies and serum adalimumab concentrations in
rheumatoid arthritis. Ann. Rheum. Dis. 66, 921926
(2007).
44. Bender, N. K. et al. Immunogenicity, efficacy and
adverse events of adalimumab in RA patients.
Rheumatol. Int. 27, 269274 (2007).
45. Hwang, W. Y. & Foote, J. Immunogenicity of
engineered antibodies. Methods 36, 310 (2005).
46. Saif, M. W. & Cohenuram, M. Role of panitumumab
in the management of metastatic colorectal cancer.
Clin. Colorectal Cancer 6, 118124 (2006).
47. Saif, M. W., Peccerillo, J. & Potter, V. Successful
re-challenge with panitumumab in patients who
developed hypersensitivity reactions to cetuximab:
report of three cases and review of literature.
Cancer Chemother. Pharmacol. 63, 10171022
(2009).
48. Chung, C. H. et al. Cetuximab-induced anaphylaxis
and IgE specific for galactose--1,3-galactose. N. Engl.
J. Med. 358, 11091117 (2008).
49. Lecluse, L. L. A. et al. Extent and clinical consequences
of antibody formation against adalimumab in patients
with plaque psoriasis. Arch. Dermatol. 146, 127132
(2010).
50. Weinblatt, M. E. et al. Adalimumab, a fully human
anti-tumor necrosis factor monoclonal antibody,
for the treatment of rheumatoid arthritis in patients
taking concomitant methotrexate: the ARMADA trial.
Arthritis Rheum. 48, 3545 (2003).
51. Nechansky, A. HAHA nothing to laugh about.
Measuring the immunogenicity (human anti-human
antibody response) induced by humanized
monoclonal antibodies applying ELISA and SPR
technology. J. Pharm. Biomed. Anal. 51, 252254
(2009).
52. Lofgren, J. A. et al. Comparing ELISA and surface
plasmon resonance for assessing clinical
immunogenicity of panitumumab. J. Immunol.
178, 74677472 (2007).
53. Jefferis, R. & LeFranc, M.-P. Human immunoglobulin
allotypes possible implications for immunogenicity.
MAbs 1, 332338 (2009).
54. Gilles, J. G. et al. Natural autoantibodies and anti-
idiotypes. Semin. Thromb. Hemost. 26, 151155
(2000).
55. Emmi, L. The role of intravenous immunoglobulin
therapy in autoimmune and inflammatory disorders.
Neurol. Sci. 23 (Suppl. 1), 18 (2002).
56. Harding, F. A. et al. The immunogenicity of humanized
and fully human antibodies: residual immunogenicity
residues in the CDR regions. Mabs 2, 256265
(2010).
57. Shankar, G., Pendley, C. & Stein, K. E. A risk-based
bioanalytical strategy for the assessment of antibody
immune responses against biological drugs. Nature
Biotech. 25, 555561 (2007).
The authors declare no competing financial interests.
FURTHER INFORMATION
tufts center for the study of Drug Development:
http://csdd.tufts.edu
All liNks Are AcTive iN THe oNliNe PDF
PersPecti ves
nrd_3229_oct10.indd 774 17/9/10 11:50:15
If it were not for the great variability
among individuals, medicine might
as well be a science and not an art.
Sir William Osler, 1892
Patients and society expect that the benefits
of licensed drugs will outweigh their risks.
Drug regulations in all major jurisdictions
are aiming to ensure a positive benefitrisk
balance. However, regulatory decisions are
based on population-level information, with
an understanding that the benefitrisk will
not necessarily be positive for all treated
patients. Patients are not equally responsive
to beneficial effects, and not equally sus-
ceptible to adverse effects. Drug developers,
drug regulators and health-care profession-
als have been aware of patient heterogeneity
and that one size doesnt fit all. Regulators
attempt to direct prescribers to appropri-
ate patient subpopulations and treatment
scenarios, whereas health-care professionals
aim to tailor drug regimens to their indi-
vidual patients, often based on phenotypic
markers that are expected to influence a
drugs pharmacokinetics or pharmacody-
namics such as creatinine clearance, disease
stage or comorbidities. In some cases, pre-
scribers can tailor doses to individuals by
titrating to their response to thedrug.
In spite of such efforts, drugs often do
not perform as well in clinical practice as in
the clinical trials that provide the basis for
marketing authorization. This difference
in benefitrisk has been termed the effi-
cacyeffectiveness gap, in which efficacy is
defined as the extent to which an interven-
tion does more good than harm under ideal
circumstances (that is, under clinical trial
conditions), whereas effectiveness is the
extent to which an intervention does more
good than harm when provided under the
usual circumstances of health care practice
(REF.1). Others have referred to efficacy as
can it work? and effectiveness as does it
work? (REF.2).
There have been several instances in
which the quality of prescription drugs dif-
fered markedly between the clinical trial
medicine assessed by regulators and that
being marketed. For example, some marketed
heparin products became contaminated with
oversulphated chondroitin sulphate
3
, and the
benefitrisk of nelfinavir (Viracept; Pfizer)
came under intense scrutiny when batches
of the medicine became contaminated with a
known genotoxic substance
4
. However, cases
in which pharmaceutical quality issues give
rise to differences between efficacy and effec-
tiveness are rare and represent a distinct set
of regulatory problems; therefore they are not
further discussedhere.
Another reason for an apparent difference
between efficacy and effectiveness may be the
inadequate follow up of minor safety signals
that arise during pre-licensing develop-
ment. We re-emphasise the responsibility of
manufacturers and regulators to ensure thor-
ough post-licensing investigation of safety
signals; it is hoped that the introduction
of Risk Management Plans (RMPs) in the
European Union (EU) and Risk Evaluation
and Mitigation Strategies (REMSs) in the
United States a few years ago may success-
fully address this important issue
5
. However,
in this Opinion article we argue that the
efficacyeffectiveness gap is, in most cases, a
problem of variability in drug response. We
describe biological and behavioural sources
of variability. We then summarize current
attempts to bridge the efficacyeffectiveness
gap by reducing variability. Finally, we specu-
late on new opportunities and technologies
in this area and what regulators can do to
advance the best use ofdrugs.
Variability and treatment scenarios
We observe that over the past decade there
were few, if any, drugs that came under close
scrutiny for their benefitrisk profile or
that had to be withdrawn from the market
OPI NI ON
Bridging the efficacyeffectiveness
gap: a regulators perspective
on addressing variability of drug
response
Hans-Georg Eichler, Eric Abadie, Alasdair Breckenridge, Bruno Flamion,
Lars L.Gustafsson, Hubert Leufkens, Malcolm Rowland, Christian K.Schneider
and Brigitte Bloechl-Daum
Abstract | Drug regulatory agencies should ensure that the benefits of drugs
outweigh their risks, but licensed medicines sometimes do not perform as
expected in everyday clinical practice. Failure may relate to lower than anticipated
efficacy or a higher than anticipated incidence or severity of adverse effects. Here
we show that the problem of benefitrisk is to a considerable degree a problem of
variability in drug response. We describe biological and behavioural sources of
variability and how these contribute to the long-known efficacyeffectiveness gap.
In this context, efficacy describes how a drug performs under conditions of clinical
trials, whereas effectiveness describes how it performs under conditions of
everyday clinical practice. We argue that a broad range of pre- and post-licensing
technologies will need to be harnessed to bridge the efficacyeffectiveness gap.
Successful approaches will not be limited to the current notion of pharmaco-
genomics-based personalized medicines, but will also entail the wider use of
electronic health-care tools to improve drug prescribing and patient adherence.
PERSPECTIVES
nrd_3501_jul11.indd 495 20/06/2011 16:34
are likely to be less immunogenic than those
with rodent-derived sequences. There is
considerable unmet medical need in the
three main areas of study for investigational
human mAbs cancer, immunological and
infectious diseases and these emerging
agents could therefore provide valuable new
treatment options.
Aaron L. Nelson was previously at the Tufts University
School of Medicine, Boston, Massachusetts 02118, USA.
Present address: Novartis Institutes for Biomedical
Research, Room 7226, 7th floor, 300 Technology
Square, Cambridge, Massachusetts 02139, USA.
Eugen Dhimolea is at the Tufts University
School of Medicine, 136 Harrison Avenue, Boston,
Janice M. Reichert is at the Tufts Center for the Study
of Drug Development, Suite 1100, 75 Kneeland Street,
Boston, Massachusetts 02111, USA.
Correspondence to J.M.R.
email: janice.reichert@tufts.edu
doi:10.1038/nrd3229
Published online 3 september 2010
1. Reichert, J. M. Monoclonal antibodies as innovative
therapeutics. Curr. Pharm. Biotechnol. 9, 423430
(2008).
2. Reichert, J. M., Rosensweig, C. J., Faden, L. B. &
Dewitz, M. C. Monoclonal antibody successes in the
clinic. Nature Biotech. 23, 10731078 (2005).
3. Reichert, J. M. Antibodies to watch in 2010. MAbs
2, 84100 (2010).
4. James, K. & Bell, G. T. Human monoclonal antibody
production. Current status and future prospects.
J. Immunol. Methods 100, 540 (1987).
5. Olsson, L. & Kaplan, H. S. Humanhuman hybridomas
producing monoclonal antibodies of predefined
antigenic specificity. Proc. Natl Acad. Sci. USA 77,
54295431 (1980).
6. Shoenfeld, Y. et al. Production of autoantibodies
by humanhuman hybridomas. J. Clin. Invest. 70,
205208 (1982).
7. Olsson, L. et al. Antibody producing humanhuman
hybridomas. II. Derivation and characterization
of an antibody specific for human leukemia cells.
J. Exp. Med. 159, 537550 (1984).
8. Kozbor, D. & Roder, J. C. Requirements for the
establishment of high-titered human monoclonal
antibodies against tetanus toxoid using the
EpsteinBarr virus technique. J. Immunol. 127,
12751280 (1981).
9. Kozbor, D., Lagarde, A. E. & Roder, J. C.
Human hybridomas constructed with antigen-specific
EpsteinBarr virus-transformed cell lines. Proc. Natl
Acad. Sci. USA 79, 66516655 (1982).
10. Beerli, R. R. & Rader, C. Mining human antibody
repertoires. MAbs 2, 361374 (2010).
11. Teng, N. N. et al. Protection against Gram-negative
bacteremia and endotoxemia with human
monoclonal IgM antibodies. Proc. Natl Acad. Sci.
USA 82, 17901794 (1985).
12. Ziegler, E. J. et al. Treatment of Gram-negative
bacteremia and septic shock with HA-1A human
monoclonal antibody against endotoxin. A
The HA-1A Sepsis Study Group. N. Engl. J. Med.
324, 429436 (1991).
13. Cross, A. S. Antiendotoxin antibodies: a dead end?
Ann. Intern. Med. 121, 5860 (1994).
14. Luce, J. M. Introduction of new technology into critical
care practice: a history of HA-1A human monoclonal
antibody against endotoxin. Crit. Care Med. 21,
12331241 (1993).
15. McCloskey, R. V., Straube, R. C., Sanders, C.,
Smith, S. M. & Smith, C. R. Treatment of septic shock
with human monoclonal antibody HA-1A. A
CHESS Trial Study Group. Ann. Intern. Med. 121, 15
(1994).
16. Abbott. Abbott Annual Report 2008. Abbott website
[online], http://www.abbott.com/static/content/microsite/
annual_report/2008/16_review1.html (2008).
17. US Food and Drug Administration. Vectibix
Panitumumab Injectable. Application No.: 125147.
Medical Review(s). FDA website [online],
http://www.accessdata.fda.gov/drugsatfda_docs/
nda/2006/125147s0000_MedR.pdf (2006).
18. Van Cutsem, E. et al. Open-label phase III trial of
panitumumab plus best supportive care compared
with best supportive care alone in patients with
chemotherapy-refractory metastatic colorectal cancer.
J. Clin. Oncol. 25, 16581664 (2007).
19. Mazumdar, S. & Greenwald, D. Golimumab. MAbs
1, 422431 (2009).
J. Med. 360, 24162425 (2009).
21. Dhimolea, E. Canakinumab. MAbs 2, 313 (2010).
22. Cingoz, O. Ustekinumab. MAbs 1, 216221 (2009).
23. Teeling, J. L. et al. The biological activity of human
epitopes on CD20. J. Immunol. 177, 362371 (2006).
24. Glennie, M. J., French, R. R., Cragg, M. S. & Taylor, R. P.
Mechanisms of killing by anti-CD20 monoclonal
antibodies. Mol. Immunol. 44, 38233837 (2007).
25. Zhang, B. Ofatumumab. MAbs 1, 326331 (2009).
26. Pageau, S. C. Denosumab. MAbs 1, 210215 (2009).
27. Mazumdar, S. Raxibacumab. MAbs 1, 531538
(2009).
28. DallEra, M. & Wofsy, D. Connective tissue diseases:
belimumab for systemic lupus erythematosus: breaking
through? Nature Rev. Rheumatol. 6, 124125 (2010).
29. Wallace, D. J. et al. A phase II, randomized, double-
blind, placebo-controlled, dose-ranging study of
belimumab in patients with active systemic lupus
erythematosus. Arthritis Rheum. 61, 11681178
(2009).
30. Hodi, F. S. et al. Improved survival with ipilimumab in
patients with metastatic melanoma. N. Engl. J. Med.
363, 711723 (2010).
31. Reichert, J. M. & Dewitz, M. C. Anti-infective
monoclonal antibodies: perils and promise of
development. Nature Rev. Drug Discov. 5, 191195
(2006).
32. Reichert, J. M. & Valge-Archer, V. E. Development
trends for monoclonal antibody cancer therapeutics.
33. Green, L. L. et al. Antigen-specific human monoclonal
antibodies from mice engineered with human Ig
heavy and light chain YACs. Nature Genet. 7, 1321
(1994).
34. Lonberg, N. et al. Antigen-specific human antibodies
from mice comprising four distinct genetic
modifications. Nature 368, 856859 (1994).
35. Green, L. L. Antibody engineering via genetic
engineering of the mouse: XenoMouse strains are a
vehicle for the facile generation of therapeutic human
monoclonal antibodies. J. Immunol. Methods 231,
1123 (1999).
36. Lonberg, N. Human antibodies from transgenic
animals. Nature Biotech. 23, 11171125 (2005).
37. McCafferty, J., Griffiths, A. D., Winter, G. & Chiswell,
D. J. Phage antibodies: filamentous phage displaying
antibody variable domains. Nature 348, 552554
(1990).
38. Vaughan, T. J. et al. Human antibodies with
sub-nanomolar affinities isolated from a large
non-immunized phage display library. Nature Biotech.
14, 309314 (1996).
39. Clackson, T., Hoogenboom, H. R., Griffiths, A. D. &
Winter, G. Making antibody fragments using phage
display libraries. Nature 352, 624628 (1991).
40. Chirino, A. J., Ary, M. L. & Marshall, S. A.
Minimizing the immunogenicity of protein
therapeutics. Drug Discov. Today 9, 8290 (2004).
41. Baert, F. et al. Influence of immunogenicity on the
long-term efficacy of infliximab in Crohns disease.
N. Engl. J. Med. 348, 601608 (2003).
42. De Groot, A. S. & Scott, D. W. Immunogenicity of
protein therapeutics. Trends Immunol. 28, 482490
(2007).
43. Bartelds, G. M. et al. Clinical response to
adalimumab: relationship to anti-adalimumab
antibodies and serum adalimumab concentrations in
rheumatoid arthritis. Ann. Rheum. Dis. 66, 921926
(2007).
44. Bender, N. K. et al. Immunogenicity, efficacy and
adverse events of adalimumab in RA patients.
Rheumatol. Int. 27, 269274 (2007).
45. Hwang, W. Y. & Foote, J. Immunogenicity of
engineered antibodies. Methods 36, 310 (2005).
46. Saif, M. W. & Cohenuram, M. Role of panitumumab
in the management of metastatic colorectal cancer.
Clin. Colorectal Cancer 6, 118124 (2006).
47. Saif, M. W., Peccerillo, J. & Potter, V. Successful
re-challenge with panitumumab in patients who
developed hypersensitivity reactions to cetuximab:
report of three cases and review of literature.
Cancer Chemother. Pharmacol. 63, 10171022
(2009).
48. Chung, C. H. et al. Cetuximab-induced anaphylaxis
and IgE specific for galactose--1,3-galactose. N. Engl.
J. Med. 358, 11091117 (2008).
49. Lecluse, L. L. A. et al. Extent and clinical consequences
of antibody formation against adalimumab in patients
with plaque psoriasis. Arch. Dermatol. 146, 127132
(2010).
50. Weinblatt, M. E. et al. Adalimumab, a fully human
anti-tumor necrosis factor monoclonal antibody,
for the treatment of rheumatoid arthritis in patients
taking concomitant methotrexate: the ARMADA trial.
Arthritis Rheum. 48, 3545 (2003).
51. Nechansky, A. HAHA nothing to laugh about.
Measuring the immunogenicity (human anti-human
antibody response) induced by humanized
monoclonal antibodies applying ELISA and SPR
technology. J. Pharm. Biomed. Anal. 51, 252254
(2009).
52. Lofgren, J. A. et al. Comparing ELISA and surface
plasmon resonance for assessing clinical
immunogenicity of panitumumab. J. Immunol.
178, 74677472 (2007).
53. Jefferis, R. & LeFranc, M.-P. Human immunoglobulin
allotypes possible implications for immunogenicity.
MAbs 1, 332338 (2009).
54. Gilles, J. G. et al. Natural autoantibodies and anti-
idiotypes. Semin. Thromb. Hemost. 26, 151155
(2000).
55. Emmi, L. The role of intravenous immunoglobulin
therapy in autoimmune and inflammatory disorders.
Neurol. Sci. 23 (Suppl. 1), 18 (2002).
56. Harding, F. A. et al. The immunogenicity of humanized
and fully human antibodies: residual immunogenicity
residues in the CDR regions. Mabs 2, 256265
(2010).
57. Shankar, G., Pendley, C. & Stein, K. E. A risk-based
bioanalytical strategy for the assessment of antibody
immune responses against biological drugs. Nature
Biotech. 25, 555561 (2007).
FURTHER INFORMATION
tufts center for the study of Drug Development:
http://csdd.tufts.edu
All liNks Are AcTive iN THe oNliNe PDF
PersPecti ves
nrd_3229_oct10.indd 774 17/9/10 11:50:15
First published in Nature Reviews Drug Discovery 10, 495-506 (July 2011); doi:10.1038/nrd3501
'Cloun', oxlunuory or ocucy riul
'Noisy', ruqmuic or oocivonoss riul
C
D
in the EU as an adjunct to diet and exercise
for the treatment of obese patients or over-
weight patients with associated risk factors
[] (REF.16). At that time, EU regulators
concluded that in clinical studies in which
the patients were treated for more than a
year, the weight reducing effect was moder-
ate and of clinical relevance for 2030% of
the treated patients [] (REF.16). There
were also safety questions from early devel-
opment; clinical trials had shown increased
rates of depressive disorders and suicidal and
aggressive behaviour, but, based on the inci-
dences observed in the trials, these adverse
effects were considered not to outweigh the
benefits. After the drug was on the market,
however, new data indicated a shorter dura-
tion of treatment in real life and, conse-
quently, a reduced beneficial effect on body
weight. At the same time, safety also became
an issue. A prescription survey revealed
that adherence by prescribers to the warn-
ings and contraindications in the label was
less than perfect. It became apparent that
the risk of experiencing the known adverse
mental effects had increased since the time
of approval, with higher absolute risks in
patients with comorbidity. Based on these
data, rimonabant was taken off the market in
the EU in 2009 (REF.16).
The rimonabant case study is instructive
because it involved a combination of some of
the causes of the efficacyeffectiveness gap
(FIG.1; TABLE1): poor patient persistence on
treatment (which reduced the effect size),
outside-label prescribing by physicians and
increased susceptibility to adverse effects
in some patients as a result of a pre-existing
comorbidity
17
. Along similar lines, prescrip-
tion of varenicline (Champix/Chantix; Pfizer)
to a broader population of patients than was
enrolled in clinical trials gave rise to height-
ened safety concerns about suicidal ideation
18
.
Sponsors of drug development pro-
grammes will understandably seek to con-
trol the treatment scenario in pre-licensing
trials in order to optimize benefitrisk.
Nonetheless, the transition from the pre-
licensing to the post-licensing stage is not
always accompanied by a deterioration of
benefitrisk. There are some cases in which
variability was reduced and benefitrisk
improved after marketing. For example, the
subgroup of patients who are treatment-
responsive to gefitinib (Iressa; AstraZeneca/
Teva) was only identified after licensing in
the United States
19
. In this case, both licens-
ing status and utilization of the drug shifted
from a broader, high-variability population
to a smaller, more circumscribed subpopula-
tion that is, from right to left in the graph
shown in FIG.1. There are other instances
in which the benefitrisk profile improved
after market introduction, including a
large number of dose adaptations (gener-
ally reductions) in the label after the initial
licensing
20,21
. These cases may be indicative
of suboptimal pre-marketing development
strategies, such as incomplete dose-ranging
studies, and the improved benefitrisk is
a result of additional research, serendipity
and/or safety-motivated regulatory action;
they do not contradict the broader notion of
increasing variability of drug response dur-
ing the post-marketingphase.
Sources of variability in drug response
Awareness of the variability in drug response
has received a boost, under the head-
ing of personalized medicine, by growing
knowledge about the genomic basis of drug
response. For some drugs, a substantial frac-
tion of variability in efficacy or toxicity can
now be explained on the basis of a single
genomic marker. Although the notion of
genomic treatment stratification has already
proved useful and personalized medicine is
expected to make significant contributions
to improving the benefitrisk of drugs, it is
unrealistic to assume that even the best pre-
dictive biomarkers will eliminate variability
or eliminate the efficacyeffectiveness gap.
This is because genetics is merely one factor
driving variability, albeit an important one.
We take a broader view that encompasses
the full range of biological and behavioural
sources of variability, as depicted in TABLE1.
In the following section, we will briefly
describe and illustrate by examples indi-
vidual sources of variability.
Biology
Pharmacogenomics. It is reasonable to
assume that differences in individual
patients genetic make-up are a major source
of variability for many drugs. It is merely our
current lack of understanding that precludes
us from exploiting these differences to better
target drugs to appropriate patients in appro-
priate doses (for an overview of pharmaco-
genomics, please refer to REF.22).
Probably the best examples to illustrate
the power of genomic markers to reduce
variability and optimize a drugs ben-
efits or risks are trastuzumab (Herceptin;
Genentech) and abacavir (Ziagen;
GlaxoSmithKline).
Trastuzumb is a monoclonal antibody
authorized in the EU for the treatment
of patients with HER2 positive meta-
static breast cancer (REF.23). Trastuzumb
targets HER2 (also known as ERBB2), a
member of the epidermal growth factor
receptor (EGFR) family of tyrosine kinase
receptors that is overexpressed in about
2530% of invasive ductal breast cancers
(HER2-positive breast cancer). Identification
of the biomarker HER2 and the develop-
ment of sufficiently reliable laboratory tests
to distinguish between HER2-positive and
HER2-negative population strata have ena-
bled the developers of this drug to focus
on the high-responder subgroup, thereby
reducing the number of patients in pre-
marketing clinical trials, as well as clinical
development cost and time. It has been spec-
ulated that without pre-selection of patients
with HER2-positive breast cancer, too many
patients and too much follow-up time would
have been needed to demonstrate a statisti-
cally significant benefit with trastuzumab
19
;
the relatively high therapeutic success rate
in the HER2-positive stratum would have
been diluted by the larger number of non-
responders in an unselected high-variability
population. Moreover, the availability of a
biomarker assay that can be implemented
in most health-care settings was shown to
adequately guide clinical practice; there is a
Figure 2 | Signal-to-noise ratio in clinical trials.
Detecting a statistically significant difference in
outcome between treatment groups (for example,
between experimental drug groups and placebo
groups) can be conceptualized as a signal-to-
noise problem. The higher the signal-to-noise
ratio, the better the chances of detecting a true
difference of a given magnitude and with a prede-
fined sample size. a | The signal-to-noise ratio in a
clinical trial with narrow patient selection criteria
and a tightly controlled treatment scenario, such
as clear definition of allowable concomitant medi-
cation and monitoring of patient adherence. This
clean setting is called an explanatory or efficacy
trial, and the intergroup difference can be
detected with ease. b | The signal-to-noise ratio of
a pragmatic or effectiveness trial; in contrast to a,
the trial population is more heterogeneous and
the treatment scenario resembles everyday clini-
cal practice. The signal may now be lost in the
noise, even though the drug has not lost its phar-
macological activity in those patients who are
responsive and adherent; hence the signal is still
there (orange oval) but is too small to detect.
PERSPECTI VES
nrd_3501_jul11.indd 497 20/06/2011 16:34
C
u
m
u
l
u
i
v
o

o
l

r
o
u
o
d

i
o
n
s

o
x
o
r
i
o
n
c
i
n
q

b
o
n
o
s

u
n
d

r
i
s
l
s
Hiqb
Vuriubiliy
Lov
Clinicul riul
sconurio
8onos
100
0
Pisls
Auborizod lubol
sconurio
Lcucy-oocivonoss qu
Avoruqo rosonsivonoss o bonociul oocs
Hiqb Lov
Avoruqo suscoibiliy o udvorso oocs
Lov Hiqb
Ousido-lubol
sconurio
Posiivo
8ono-risl
Noquivo
8ono-risl
Vuriubiliy
C
D
but that also failed to be beneficial for at
least some patient subgroups. It has been
speculated that this assumption may even
be true for rofecoxib (Vioxx; Merck)
6
,
which has triggered much debate about
the adequacy of drug licensing. Failure of a
licensed drug across the entire treatment-
eligible population would be very surpris-
ing indeed, considering the ever tighter
and more sophisticated evidentiary and
analytical standards applied to drug licens-
ing decisions. Past experience shows that it
is usually only a small subgroup of treated
patients who experience debilitating or life-
threatening adverse effects, and, likewise,
only a small subgroup who do not respond
to treatment. However, it should be noted
that the fraction of non-responders may be
as high as 3060% in some cases
7,8
.
To make matters even more complex,
drug response and benefitrisk profile are
often not black and white but are shades
of grey. A useful example of the gradient
of benefitrisk across patient strata was
provided nearly 15years ago, based on pre-
and post-licensing data on statin treatment.
Pharoah and Hollingworth
9
identified more
than 40 different patient strata, constructed
from a combination of demographic and
phenotypic criteria including age, sex, previ-
ous myocardial infarction and cholesterol
level. Estimates of clinical efficacy and cost-
effectiveness of the lipid-lowering treatments
provided in their analysis varied 15-fold
between riskstrata.
Such findings are expected, as the benefits
and risks from drugs vary from one patient
to the next, and are dependent on the
treatment scenario (FIG.1). Regulatory deci-
sions are taken on the basis of results from
randomized controlled trials (RCTs) in
which patient enrolment is restricted by
long lists of inclusion and exclusion crite-
ria. In addition, dose regimens, including
concomitant medications, are specified in
the trial protocol; trial staff make a special
effort to convince patients to take the trial
medications according to the prescribed
dosing regimen, and safety is monitored
at pre-specified intervals. These measures
are taken in the hope of optimizing treat-
ment conditions and selecting homogenous
patient groups with a high responsiveness
for the desired pharmacologic effect and a
low susceptibility for toxicity (for example,
by eliminating patients with some pre-
existing comorbidities). Thereby, variability
is minimized and the signal-to-noise ratio
enhanced (FIG.2). This enables drug develop-
ers to demonstrate a positive benefitrisk
profile. If regulators license the drug, it will
often be for a less-well-described patient
population. The label normally lists fewer
constraints than the clinical trial protocols
for dose regimens, age-range, comorbid
conditions or concomitant drug treatments.
Thus, even if prescribers were to strictly
adhere to the label, variability is expected to
be higher in the label scenario than in the
RCT scenario, and the benefitrisk may be
reduced (FIG.1).
Studies of drug utilization in daily prac-
tice have shown that drugs will frequently
be prescribed or taken for non-approved
conditions or otherwise not in accordance
with the label
1012
, and real-life treatment
scenarios are characterized by even wider
variability than on-label.
To a large extent, the efficacyeffectiveness
gap may be considered a result of increas-
ing variability of drug response owing to
a combination of genetic, other biological
and behavioural factors, as summarized
in TABLE1 and described below. Increased
variability may make the benefitrisk of a
drug look less favourable, calling into ques-
tion the regulatory licensing decision and in
some cases necessitating regulatory action.
At least two drugs, cerivastatin (Lipobay;
Bayer) and mibefradil (Posicor; Roche), had
to be taken off the market, partly as a result
of inappropriate prescribing. Both drugs
appeared to have a favourable benefitrisk at
the time of licensing, but recommendations
and warnings in the label were ignored by
prescribers
1315
.
A recent example of the efficacy
effectiveness gap is rimonabant (Acomplia;
Sanofi-Aventis), which was licensed in 2006
Figure 1 | Average benefitrisk of drugs as a function of treatment scenario. a | In the context
of regulatory clinical trials (left-hand side), variability in drug response is deliberately kept to a mini-
mum by enforced treatment conditions and narrow selection criteria, which aim to restrict the patient
population to high-responders and good-toleraters. Criteria for treatment eligibility and treatment
scenarios are usually less constrained in the authorized drug label (middle), and even less so after the
drug comes to market and is partly prescribed and taken outside the label scenario (different popula-
tions, doses, patients with contraindications and comorbidities, poor adherence, and so on; right-hand
side). As variability in drug response increases from left to right (red line), average benefitrisk deterio-
rates (blue line), giving rise to the efficacyeffectiveness gap. b | The deterioration in average benefit
risk is a result of the diminishing responsiveness to the beneficial effects (red line) and the increasing
susceptibility to adverse drug effects (blue line), as more and different patient strata are added. Note
that responsiveness to beneficial drug effects or susceptibility to adverse drug effects are not merely
determined genetically but also by environmental and behavioural factors (for example, inappropriate
prescribing or poor patient adherence to treatment regimens).
PERSPECTI VES
nrd_3501_jul11.indd 496 20/06/2011 16:34
'Cloun', oxlunuory or ocucy riul
'Noisy', ruqmuic or oocivonoss riul
C
D
in the EU as an adjunct to diet and exercise
for the treatment of obese patients or over-
weight patients with associated risk factors
[] (REF.16). At that time, EU regulators
concluded that in clinical studies in which
the patients were treated for more than a
year, the weight reducing effect was moder-
ate and of clinical relevance for 2030% of
the treated patients [] (REF.16). There
were also safety questions from early devel-
opment; clinical trials had shown increased
rates of depressive disorders and suicidal and
aggressive behaviour, but, based on the inci-
dences observed in the trials, these adverse
effects were considered not to outweigh the
benefits. After the drug was on the market,
however, new data indicated a shorter dura-
tion of treatment in real life and, conse-
quently, a reduced beneficial effect on body
weight. At the same time, safety also became
an issue. A prescription survey revealed
that adherence by prescribers to the warn-
ings and contraindications in the label was
less than perfect. It became apparent that
the risk of experiencing the known adverse
mental effects had increased since the time
of approval, with higher absolute risks in
patients with comorbidity. Based on these
data, rimonabant was taken off the market in
the EU in 2009 (REF.16).
The rimonabant case study is instructive
because it involved a combination of some of
the causes of the efficacyeffectiveness gap
(FIG.1; TABLE1): poor patient persistence on
treatment (which reduced the effect size),
outside-label prescribing by physicians and
increased susceptibility to adverse effects
in some patients as a result of a pre-existing
comorbidity
17
. Along similar lines, prescrip-
tion of varenicline (Champix/Chantix; Pfizer)
to a broader population of patients than was
enrolled in clinical trials gave rise to height-
ened safety concerns about suicidal ideation
18
.
Sponsors of drug development pro-
grammes will understandably seek to con-
trol the treatment scenario in pre-licensing
trials in order to optimize benefitrisk.
Nonetheless, the transition from the pre-
licensing to the post-licensing stage is not
always accompanied by a deterioration of
benefitrisk. There are some cases in which
variability was reduced and benefitrisk
improved after marketing. For example, the
subgroup of patients who are treatment-
responsive to gefitinib (Iressa; AstraZeneca/
Teva) was only identified after licensing in
the United States
19
. In this case, both licens-
ing status and utilization of the drug shifted
from a broader, high-variability population
to a smaller, more circumscribed subpopula-
tion that is, from right to left in the graph
shown in FIG.1. There are other instances
in which the benefitrisk profile improved
after market introduction, including a
large number of dose adaptations (gener-
ally reductions) in the label after the initial
licensing
20,21
. These cases may be indicative
of suboptimal pre-marketing development
strategies, such as incomplete dose-ranging
studies, and the improved benefitrisk is
a result of additional research, serendipity
and/or safety-motivated regulatory action;
they do not contradict the broader notion of
increasing variability of drug response dur-
ing the post-marketingphase.
Sources of variability in drug response
Awareness of the variability in drug response
has received a boost, under the head-
ing of personalized medicine, by growing
knowledge about the genomic basis of drug
response. For some drugs, a substantial frac-
tion of variability in efficacy or toxicity can
now be explained on the basis of a single
genomic marker. Although the notion of
genomic treatment stratification has already
proved useful and personalized medicine is
expected to make significant contributions
to improving the benefitrisk of drugs, it is
unrealistic to assume that even the best pre-
dictive biomarkers will eliminate variability
or eliminate the efficacyeffectiveness gap.
This is because genetics is merely one factor
driving variability, albeit an important one.
We take a broader view that encompasses
the full range of biological and behavioural
sources of variability, as depicted in TABLE1.
In the following section, we will briefly
describe and illustrate by examples indi-
vidual sources of variability.
Biology
Pharmacogenomics. It is reasonable to
assume that differences in individual
patients genetic make-up are a major source
of variability for many drugs. It is merely our
current lack of understanding that precludes
us from exploiting these differences to better
target drugs to appropriate patients in appro-
priate doses (for an overview of pharmaco-
genomics, please refer to REF.22).
Probably the best examples to illustrate
the power of genomic markers to reduce
variability and optimize a drugs ben-
efits or risks are trastuzumab (Herceptin;
Genentech) and abacavir (Ziagen;
GlaxoSmithKline).
Trastuzumb is a monoclonal antibody
authorized in the EU for the treatment
of patients with HER2 positive meta-
static breast cancer (REF.23). Trastuzumb
targets HER2 (also known as ERBB2), a
member of the epidermal growth factor
receptor (EGFR) family of tyrosine kinase
receptors that is overexpressed in about
2530% of invasive ductal breast cancers
(HER2-positive breast cancer). Identification
of the biomarker HER2 and the develop-
ment of sufficiently reliable laboratory tests
to distinguish between HER2-positive and
HER2-negative population strata have ena-
bled the developers of this drug to focus
on the high-responder subgroup, thereby
reducing the number of patients in pre-
marketing clinical trials, as well as clinical
development cost and time. It has been spec-
ulated that without pre-selection of patients
with HER2-positive breast cancer, too many
patients and too much follow-up time would
have been needed to demonstrate a statisti-
cally significant benefit with trastuzumab
19
;
the relatively high therapeutic success rate
in the HER2-positive stratum would have
been diluted by the larger number of non-
responders in an unselected high-variability
population. Moreover, the availability of a
biomarker assay that can be implemented
in most health-care settings was shown to
adequately guide clinical practice; there is a
Figure 2 | Signal-to-noise ratio in clinical trials.
Detecting a statistically significant difference in
outcome between treatment groups (for example,
between experimental drug groups and placebo
groups) can be conceptualized as a signal-to-
noise problem. The higher the signal-to-noise
ratio, the better the chances of detecting a true
difference of a given magnitude and with a prede-
fined sample size. a | The signal-to-noise ratio in a
clinical trial with narrow patient selection criteria
and a tightly controlled treatment scenario, such
as clear definition of allowable concomitant medi-
cation and monitoring of patient adherence. This
clean setting is called an explanatory or efficacy
trial, and the intergroup difference can be
detected with ease. b | The signal-to-noise ratio of
a pragmatic or effectiveness trial; in contrast to a,
the trial population is more heterogeneous and
the treatment scenario resembles everyday clini-
cal practice. The signal may now be lost in the
noise, even though the drug has not lost its phar-
macological activity in those patients who are
responsive and adherent; hence the signal is still
there (orange oval) but is too small to detect.
PERSPECTI VES
nrd_3501_jul11.indd 497 20/06/2011 16:34
C
u
m
u
l
u
i
v
o

o
l

r
o
u
o
d

i
o
n
s

o
x
o
r
i
o
n
c
i
n
q

b
o
n
o
s

u
n
d

r
i
s
l
s
Hiqb
Vuriubiliy
Lov
Clinicul riul
sconurio
8onos
100
0
Pisls
Auborizod lubol
sconurio
Lcucy-oocivonoss qu
Avoruqo rosonsivonoss o bonociul oocs
Hiqb Lov
Avoruqo suscoibiliy o udvorso oocs
Lov Hiqb
Ousido-lubol
sconurio
Posiivo
8ono-risl
Noquivo
8ono-risl
Vuriubiliy
C
D
but that also failed to be beneficial for at
least some patient subgroups. It has been
speculated that this assumption may even
be true for rofecoxib (Vioxx; Merck)
6
,
which has triggered much debate about
the adequacy of drug licensing. Failure of a
licensed drug across the entire treatment-
eligible population would be very surpris-
ing indeed, considering the ever tighter
and more sophisticated evidentiary and
analytical standards applied to drug licens-
ing decisions. Past experience shows that it
is usually only a small subgroup of treated
patients who experience debilitating or life-
threatening adverse effects, and, likewise,
only a small subgroup who do not respond
to treatment. However, it should be noted
that the fraction of non-responders may be
as high as 3060% in some cases
7,8
.
To make matters even more complex,
drug response and benefitrisk profile are
often not black and white but are shades
of grey. A useful example of the gradient
of benefitrisk across patient strata was
provided nearly 15years ago, based on pre-
and post-licensing data on statin treatment.
Pharoah and Hollingworth
9
identified more
than 40 different patient strata, constructed
from a combination of demographic and
phenotypic criteria including age, sex, previ-
ous myocardial infarction and cholesterol
level. Estimates of clinical efficacy and cost-
effectiveness of the lipid-lowering treatments
provided in their analysis varied 15-fold
between riskstrata.
Such findings are expected, as the benefits
and risks from drugs vary from one patient
to the next, and are dependent on the
treatment scenario (FIG.1). Regulatory deci-
sions are taken on the basis of results from
randomized controlled trials (RCTs) in
which patient enrolment is restricted by
long lists of inclusion and exclusion crite-
ria. In addition, dose regimens, including
concomitant medications, are specified in
the trial protocol; trial staff make a special
effort to convince patients to take the trial
medications according to the prescribed
dosing regimen, and safety is monitored
at pre-specified intervals. These measures
are taken in the hope of optimizing treat-
ment conditions and selecting homogenous
patient groups with a high responsiveness
for the desired pharmacologic effect and a
low susceptibility for toxicity (for example,
by eliminating patients with some pre-
existing comorbidities). Thereby, variability
is minimized and the signal-to-noise ratio
enhanced (FIG.2). This enables drug develop-
ers to demonstrate a positive benefitrisk
profile. If regulators license the drug, it will
often be for a less-well-described patient
population. The label normally lists fewer
constraints than the clinical trial protocols
for dose regimens, age-range, comorbid
conditions or concomitant drug treatments.
Thus, even if prescribers were to strictly
adhere to the label, variability is expected to
be higher in the label scenario than in the
RCT scenario, and the benefitrisk may be
reduced (FIG.1).
Studies of drug utilization in daily prac-
tice have shown that drugs will frequently
be prescribed or taken for non-approved
conditions or otherwise not in accordance
with the label
1012
, and real-life treatment
scenarios are characterized by even wider
variability than on-label.
To a large extent, the efficacyeffectiveness
gap may be considered a result of increas-
ing variability of drug response owing to
a combination of genetic, other biological
and behavioural factors, as summarized
in TABLE1 and described below. Increased
variability may make the benefitrisk of a
drug look less favourable, calling into ques-
tion the regulatory licensing decision and in
some cases necessitating regulatory action.
At least two drugs, cerivastatin (Lipobay;
Bayer) and mibefradil (Posicor; Roche), had
to be taken off the market, partly as a result
of inappropriate prescribing. Both drugs
appeared to have a favourable benefitrisk at
the time of licensing, but recommendations
and warnings in the label were ignored by
prescribers
1315
.
A recent example of the efficacy
effectiveness gap is rimonabant (Acomplia;
Sanofi-Aventis), which was licensed in 2006
Figure 1 | Average benefitrisk of drugs as a function of treatment scenario. a | In the context
of regulatory clinical trials (left-hand side), variability in drug response is deliberately kept to a mini-
mum by enforced treatment conditions and narrow selection criteria, which aim to restrict the patient
population to high-responders and good-toleraters. Criteria for treatment eligibility and treatment
scenarios are usually less constrained in the authorized drug label (middle), and even less so after the
drug comes to market and is partly prescribed and taken outside the label scenario (different popula-
tions, doses, patients with contraindications and comorbidities, poor adherence, and so on; right-hand
side). As variability in drug response increases from left to right (red line), average benefitrisk deterio-
rates (blue line), giving rise to the efficacyeffectiveness gap. b | The deterioration in average benefit
risk is a result of the diminishing responsiveness to the beneficial effects (red line) and the increasing
susceptibility to adverse drug effects (blue line), as more and different patient strata are added. Note
that responsiveness to beneficial drug effects or susceptibility to adverse drug effects are not merely
determined genetically but also by environmental and behavioural factors (for example, inappropriate
prescribing or poor patient adherence to treatment regimens).
PERSPECTI VES
nrd_3501_jul11.indd 496 20/06/2011 16:34
shown in many instances to contribute to the
efficacyeffectiveness gap owing to more fre-
quent adverse effects or reduced benefit
3638
.
Off-label prescribing is expected to increase
variability of drug response, for example
because new patients with pre-existing
morbidities are being added to the treat-
ment population, as occurred in the cases of
rimonabant and varenicline (see above).
Effectiveness may be further reduced by
inadvertent medication errors, including dis-
pensing of the wrong drug or dose strength,
inappropriate handling and storage of drugs,
and other more subtle examples of poor use
of medicines that may occur throughout the
medication use system. The impact on pub-
lic health is huge and studies have estimated
that medication errors, including inappro-
priate prescribing, cost tens of thousands of
lives each year in the United Statesalone
39
.
Patient adherence. Drugs dont work in
patients who dont take them. Also known
as Koops Law, that simple observation
by C. Everett Koop, former US surgeon
general
40
, was the subject of numerous stud-
ies and even of a dedicated World Health
Organization report
41
. Poor adherence (also
termed compliance) by patients to pre-
scribed drug regimens may come in many
forms, ranging from fluctuations in dose-
timing to omitting a single days dosing to
having repeat drug holidays (the sequential
omission of 3 or more days doses) and,
finally, to non-persistence (not taking drugs
prescribed for chronic conditions)
42
.
Patients adherence is not perfect even
in pre-licensing trials, and especially in
long-term outcome trials
43
. Nonetheless,
regulatory trials are usually geared towards
maintaining adherence at the highest
possible level, and adherence in real-life
populations is expected to be substantially
lower than in clinical trials
44
. It has been
suggested that poor adherence may be the
single largest source of variance in drug
responses in some settings
45
, and it is also
likely to be one of the principal sources of
the efficacyeffectivenessgap.
There is an abundance of reports docu-
menting that poor adherence or non-adher-
ence by patients is a cause of potentially
avoidable morbidity, mortality and lost pro-
ductivity. The one class of drugs for which
the importance of adherence is probably best
understood and for which lack of adher-
ence has been described as the Achilles heel
is oral contraceptives. Because of their
relatively low variability in pharmacokinetics
and pharmacodynamics, these drugs could
achieve a near-perfect therapeutic success
rate that is, near zero risk of conception
given punctual dosing
46
. However, based
on one estimate, in real-life conditions the
conception rate increases 50-fold, from 0.1%
per year under perfect use conditions to
5.0% under typical use (REF.47).
Studies in HIV-infected patients have
demonstrated a significant relation between
patient adherence and virologic response
48

and possibly also drug-resistance devel-
opment
49
. In a study of patients with
schizophrenia, approximately 7% of patients
become non-adherent to their prescribed
drug regimen each month after discharge
from the hospital, and non-adherence was
associated with an increased relapse rate
50
.
Vrijens etal. reported that about half of
the patients who were prescribed an anti-
hypertensive drug had stopped taking it
within 1year
51
; it is easy to see how lack
of drug exposure due to poor adherence
may, if it is not reported by the patient and
remains undetected by the prescriber, lead
to the interpretation of treatment-resistant
hypertension. Adherence is likely to become
a factor in driving the clinical outcome of
cancer treatments as more and more cancer
drugs are administered orally and need to
be taken over prolonged periods and in an
ambulatory setting. Poor tamoxifen adher-
ence was recently shown to be significantly
associated with an increased risk of breast
cancer events
52
.
In some treatment situations, adher-
ence may take on a broader meaning than
just taking a drug dose at a given time and
include, for example, dieting and exercising.
A lack of lifestyle changes may mitigate the
therapeutic success of drug treatment. This
was shown by recent findings that patients
with early rheumatoid arthritis who smoke
are less likely to respond to treatment with
methotrexate and TNF inhibitors
53
.
Variability sources in a real-life setting
We have briefly described the individual
influences that govern variability in any
given therapeutic scenario. In everyday clini-
cal practice, all of them will influence overall
variability and the extent of the efficacy
effectiveness gap. However, the interplay of
these biological and behavioural factors is
not straightforward.
In an early seminal paper, Harter and
Peck
54
discussed major sources of variability
in doseeffect relationships and, impor-
tantly, their highly nonlinear interactions.
They emphasized that total variability is
the square root of the sum of the squares
of the variabilities of each individual
source
54
. Urquhart
46
pointed out that: this
nonlinearity has the unhappy consequence
that a major reduction in the variability of
one component of the system has only a
minor effect on the overall variability.
Harter and Peck also attempted to quan-
tify for one model drug the relative contribu-
tion of individual sources of variability to the
cumulative variability of the overall thera-
peutic response. In their estimate, adherence
and pharmacokinetics are the most impor-
tant drivers of variability of the therapeutic
effect of theophylline, a drug that was widely
used at the time in patients with asthma
54
.
However, the relative contribution of indi-
vidual sources to overall variability is likely
to be different from one drug to the next and
is unknown, or at least not formally assessed,
for mostdrugs.
Bridging the efficacyeffectiveness gap
Whenever expectations (efficacy) do not
match reality (effectiveness), there are two
ways to bridge the gap: either lower the
expectations to the level of reality (in our
context this would mean regulators demand
pre-marketing studies that fully represent
clinical reality and then base licensing deci-
sions on effectiveness information) or bring
up reality to the level of expectations. This
second approach seeks to ensure that every-
day practice strictly follows the label and that
drug regimens are individualized to meet
patients needs. Both concepts are useful but
present serious challenges.
Lower the expectations to the level of reality.
Ideally, regulatory RCTs should be inter-
nally valid (that is, their design and conduct
must minimize the possibility of bias), but
to be clinically useful, the result must also
be relevant to a definable group of patients
in a particular clinical setting (that is, they
must be externally valid)
55
. A perceived lack
of external validity is a frequent criticism
of current pre-marketing RCTs, which are
often referred to as explanatory or efficacy
trials. There is a call for trials with higher
external validity, also known as pragmatic
clinical trials (PCTs) or effectiveness trials
55
.
For an in-depth discussion of the domains
that characterize explanatory versus prag-
matic trials, please refer to the PRECIS
framework (REF.56).
An argument can indeed be made in
favour of a shift in regulatory evidence
standards from efficacy towards effective-
ness. Such a shift would imply fewer and
broader patient selection criteria (with the
exception of validated biomarker-guided
selection criteria, see below) and less control
over patient management. Pre-licensing
PERSPECTI VES
nrd_3501_jul11.indd 499 20/06/2011 16:34
documented high rate of adherence to HER2
testing and treatment recommendations
24

that may reduce the efficacyeffectiveness
gap for this drug.
Abacavir shows what pharmacogenomics
can do for drug safety. Abacavir is a nucleo-
side reverse transcriptase inhibitor of HIV-1
and HIV-2, and is indicated for antiretroviral
combination therapy of HIV infection. The
dominant toxicity of this drug is a risk of
hypersensitivity reactions. Propensity stud-
ies demonstrated that the incidence could
be dramatically reduced by pre-therapy
screening for the human leukocyte anti-
gen (HLA)B*5701 allele. It is currently
estimated that 4861% of patients with the
HLAB*5701 allele will develop a hyper-
sensitivity reaction to abacavir, compared
to only 04% of patients who do not have
the HLAB*5701 allele
25
. Detection of the
genetic marker has substantially improved
the benefitrisk of this compound by
de-selecting the majority of patients who
would experience a serious adverseeffect.
Pharmacogenomics also determines
pharmacokinetics, and the knowledge about
polymorphisms in genes regulating drug-
metabolizing enzymes or drug transporters
has been successfully exploited to adjust
drug dose regimens in an attempt to reduce
variability in drug plasma levels and, conse-
quently, in benefitrisk
26,27
.
Other intrinsic and extrinsic factors. These
sources of variability are a mixed bag of
sometimes poorly described factors that
co-determine pharmacokinetics and/or
pharmacodynamics. Non-genetic variabil-
ity factors are divided into intrinsic and
extrinsic factors. Intrinsic factors are often
referred to as different states of physiol-
ogy or pathology
28
, such as age, sex, body
weight
29
, comorbidities or baseline severity
of disease
30,31
. Extrinsic factors are environ-
mental influences
28
, such as pollution, sun-
light, co-medication or influences of food on
pharmacokinetics, such as consumption of
fruit juice
32,33
.
For instance, patients with rheumatoid
arthritis who had a high body mass index
were reported to respond less well than
patients with a low body mass index to
tumour necrosis factor (TNF) blockade
by infliximab, even when the results were
adjusted for baseline status of disease activity
and infliximab dosage was based on body
weight. As fat tissue may be modulating the
inflammatory disease process, this finding
supports the notion that a patients volume
of fat tissue could have implications for other
immune-mediated inflammatory conditions
treated with TNF antagonists
34
.
Some of the residual (or unexplained)
variability ascribed to intrinsic and extrinsic
factors might, in reality, be due to genetic or
behavioural factors but is presently not rec-
ognized as such. Hence, it will be impossible
to address these modifiers of benefitrisk,
and eliminating all biological variability of
drug response will remain an elusivegoal.
Behaviour
Prescribing and drug handling in the medi-
cation use system. Regulators approve drugs
for defined indications and conditions, but
the drugs everyday use in the medication
use system is a different matter. We refer to
the medication use system as the complex
collaboration among physicians, nurses,
pharmacists, hospital or nursing home
staff and other health-care professionals
who influence drug utilization in everyday
practice
5
.
A large fraction of all prescriptions are
written off-label, sometimes encouraged
by inappropriate drug company marketing
and promotions efforts
35
. Off-label implies
prescribing for an unapproved condition,
dose regimen, treatment duration, route of
administration or for patients who should
not be given the drug because of existing
contraindications or interacting concomitant
treatment. Some off-label uses may follow
evidence of the drugs efficacy in unapproved
conditions, or may simply happen ahead of
a regulatory label change, and as such may
benefit patients. However, off-label use was
Table 1 | Sources of variability in drug response that may cause toxicity or lack of efficacy
Source of variability Mechanism Drug therapy Description of problem
Biology
Genomics Individual patients genomic make-up
influences PK and PD, affecting drug
concentration profile at the target site and
the likelihood and magnitude of desired and
adverse effects
Trastuzumab PD: only effective in patients overexpressing HER2
receptor on tumour cells
Abacavir PD: High risk of severe hypersensitivity reaction in
patients with HLAB*5701 allele
Codeine PK: lack of analgesia in carriers of nonfunctional
CYP2D6 alleles, toxicity with multiple CYP2D6 gene
copies
Other intrinsic and
extrinsic factors
Comorbidity, baseline severity of disease,
other altered physiological states, or external
factors influencing PK and/or PD
Insulin PD: glucose control and risk of hypoglycaemia
affected by stress or physical activity
Several drugs PK: increased toxicity due to increased absorption
with concomitant consumption of grapefruit juice
Behaviour
Prescribing and drug
handling
Inappropriate or off-label prescribing,
co-prescribing with an interacting drug,
continued prescribing to non-responders or
medication errors
Cerivastatin Rhabdomyolysis due to too high starting doses
Gemfibrozil Interactions in drug label ignored
Mibefradil Toxicity due to drugdrug interactions, label often
ignored
Patient adherence Poor adherence to prescribed drug regimen,
discontinuation (non-persistence), drug
holidays or inadvertent overdosing
Anti-hypertensive
drugs
Non-adherence or non-persistence perceived as
treatment-resistant hypertension
Anti-infective
therapies
Drug holidays leading to the development of
resistance
The overall variability and the extent of the efficacyeffectiveness gap is driven by the interplay between biological and behavioural factors. CYP2D6, cytochrome
P450 2D6; HLA, human leukocyte antigen; PD, pharmacodynamics; PK, pharmacokinetics.
PERSPECTI VES
nrd_3501_jul11.indd 498 20/06/2011 16:34
shown in many instances to contribute to the
efficacyeffectiveness gap owing to more fre-
quent adverse effects or reduced benefit
3638
.
Off-label prescribing is expected to increase
variability of drug response, for example
because new patients with pre-existing
morbidities are being added to the treat-
ment population, as occurred in the cases of
rimonabant and varenicline (see above).
Effectiveness may be further reduced by
inadvertent medication errors, including dis-
pensing of the wrong drug or dose strength,
inappropriate handling and storage of drugs,
and other more subtle examples of poor use
of medicines that may occur throughout the
medication use system. The impact on pub-
lic health is huge and studies have estimated
that medication errors, including inappro-
priate prescribing, cost tens of thousands of
lives each year in the United Statesalone
39
.
Patient adherence. Drugs dont work in
patients who dont take them. Also known
as Koops Law, that simple observation
by C. Everett Koop, former US surgeon
general
40
, was the subject of numerous stud-
ies and even of a dedicated World Health
Organization report
41
. Poor adherence (also
termed compliance) by patients to pre-
scribed drug regimens may come in many
forms, ranging from fluctuations in dose-
timing to omitting a single days dosing to
having repeat drug holidays (the sequential
omission of 3 or more days doses) and,
finally, to non-persistence (not taking drugs
prescribed for chronic conditions)
42
.
Patients adherence is not perfect even
in pre-licensing trials, and especially in
long-term outcome trials
43
. Nonetheless,
regulatory trials are usually geared towards
maintaining adherence at the highest
possible level, and adherence in real-life
populations is expected to be substantially
lower than in clinical trials
44
. It has been
suggested that poor adherence may be the
single largest source of variance in drug
responses in some settings
45
, and it is also
likely to be one of the principal sources of
the efficacyeffectivenessgap.
There is an abundance of reports docu-
menting that poor adherence or non-adher-
ence by patients is a cause of potentially
avoidable morbidity, mortality and lost pro-
ductivity. The one class of drugs for which
the importance of adherence is probably best
understood and for which lack of adher-
ence has been described as the Achilles heel
is oral contraceptives. Because of their
relatively low variability in pharmacokinetics
and pharmacodynamics, these drugs could
achieve a near-perfect therapeutic success
rate that is, near zero risk of conception
given punctual dosing
46
. However, based
on one estimate, in real-life conditions the
conception rate increases 50-fold, from 0.1%
per year under perfect use conditions to
5.0% under typical use (REF.47).
Studies in HIV-infected patients have
demonstrated a significant relation between
patient adherence and virologic response
48

and possibly also drug-resistance devel-
opment
49
. In a study of patients with
schizophrenia, approximately 7% of patients
become non-adherent to their prescribed
drug regimen each month after discharge
from the hospital, and non-adherence was
associated with an increased relapse rate
50
.
Vrijens etal. reported that about half of
the patients who were prescribed an anti-
hypertensive drug had stopped taking it
within 1year
51
; it is easy to see how lack
of drug exposure due to poor adherence
may, if it is not reported by the patient and
remains undetected by the prescriber, lead
to the interpretation of treatment-resistant
hypertension. Adherence is likely to become
a factor in driving the clinical outcome of
cancer treatments as more and more cancer
drugs are administered orally and need to
be taken over prolonged periods and in an
ambulatory setting. Poor tamoxifen adher-
ence was recently shown to be significantly
associated with an increased risk of breast
cancer events
52
.
In some treatment situations, adher-
ence may take on a broader meaning than
just taking a drug dose at a given time and
include, for example, dieting and exercising.
A lack of lifestyle changes may mitigate the
therapeutic success of drug treatment. This
was shown by recent findings that patients
with early rheumatoid arthritis who smoke
are less likely to respond to treatment with
methotrexate and TNF inhibitors
53
.
Variability sources in a real-life setting
We have briefly described the individual
influences that govern variability in any
given therapeutic scenario. In everyday clini-
cal practice, all of them will influence overall
variability and the extent of the efficacy
effectiveness gap. However, the interplay of
these biological and behavioural factors is
not straightforward.
In an early seminal paper, Harter and
Peck
54
discussed major sources of variability
in doseeffect relationships and, impor-
tantly, their highly nonlinear interactions.
They emphasized that total variability is
the square root of the sum of the squares
of the variabilities of each individual
source
54
. Urquhart
46
pointed out that: this
nonlinearity has the unhappy consequence
that a major reduction in the variability of
one component of the system has only a
minor effect on the overall variability.
Harter and Peck also attempted to quan-
tify for one model drug the relative contribu-
tion of individual sources of variability to the
cumulative variability of the overall thera-
peutic response. In their estimate, adherence
and pharmacokinetics are the most impor-
tant drivers of variability of the therapeutic
effect of theophylline, a drug that was widely
used at the time in patients with asthma
54
.
However, the relative contribution of indi-
vidual sources to overall variability is likely
to be different from one drug to the next and
is unknown, or at least not formally assessed,
for mostdrugs.
Bridging the efficacyeffectiveness gap
Whenever expectations (efficacy) do not
match reality (effectiveness), there are two
ways to bridge the gap: either lower the
expectations to the level of reality (in our
context this would mean regulators demand
pre-marketing studies that fully represent
clinical reality and then base licensing deci-
sions on effectiveness information) or bring
up reality to the level of expectations. This
second approach seeks to ensure that every-
day practice strictly follows the label and that
drug regimens are individualized to meet
patients needs. Both concepts are useful but
present serious challenges.
Lower the expectations to the level of reality.
Ideally, regulatory RCTs should be inter-
nally valid (that is, their design and conduct
must minimize the possibility of bias), but
to be clinically useful, the result must also
be relevant to a definable group of patients
in a particular clinical setting (that is, they
must be externally valid)
55
. A perceived lack
of external validity is a frequent criticism
of current pre-marketing RCTs, which are
often referred to as explanatory or efficacy
trials. There is a call for trials with higher
external validity, also known as pragmatic
clinical trials (PCTs) or effectiveness trials
55
.
For an in-depth discussion of the domains
that characterize explanatory versus prag-
matic trials, please refer to the PRECIS
framework (REF.56).
An argument can indeed be made in
favour of a shift in regulatory evidence
standards from efficacy towards effective-
ness. Such a shift would imply fewer and
broader patient selection criteria (with the
exception of validated biomarker-guided
selection criteria, see below) and less control
over patient management. Pre-licensing
PERSPECTI VES
nrd_3501_jul11.indd 499 20/06/2011 16:34
documented high rate of adherence to HER2
testing and treatment recommendations
24

that may reduce the efficacyeffectiveness
gap for this drug.
Abacavir shows what pharmacogenomics
can do for drug safety. Abacavir is a nucleo-
side reverse transcriptase inhibitor of HIV-1
and HIV-2, and is indicated for antiretroviral
combination therapy of HIV infection. The
dominant toxicity of this drug is a risk of
hypersensitivity reactions. Propensity stud-
ies demonstrated that the incidence could
be dramatically reduced by pre-therapy
screening for the human leukocyte anti-
gen (HLA)B*5701 allele. It is currently
estimated that 4861% of patients with the
HLAB*5701 allele will develop a hyper-
sensitivity reaction to abacavir, compared
to only 04% of patients who do not have
the HLAB*5701 allele
25
. Detection of the
genetic marker has substantially improved
the benefitrisk of this compound by
de-selecting the majority of patients who
would experience a serious adverseeffect.
Pharmacogenomics also determines
pharmacokinetics, and the knowledge about
polymorphisms in genes regulating drug-
metabolizing enzymes or drug transporters
has been successfully exploited to adjust
drug dose regimens in an attempt to reduce
variability in drug plasma levels and, conse-
quently, in benefitrisk
26,27
.
Other intrinsic and extrinsic factors. These
sources of variability are a mixed bag of
sometimes poorly described factors that
co-determine pharmacokinetics and/or
pharmacodynamics. Non-genetic variabil-
ity factors are divided into intrinsic and
extrinsic factors. Intrinsic factors are often
referred to as different states of physiol-
ogy or pathology
28
, such as age, sex, body
weight
29
, comorbidities or baseline severity
of disease
30,31
. Extrinsic factors are environ-
mental influences
28
, such as pollution, sun-
light, co-medication or influences of food on
pharmacokinetics, such as consumption of
fruit juice
32,33
.
For instance, patients with rheumatoid
arthritis who had a high body mass index
were reported to respond less well than
patients with a low body mass index to
tumour necrosis factor (TNF) blockade
by infliximab, even when the results were
adjusted for baseline status of disease activity
and infliximab dosage was based on body
weight. As fat tissue may be modulating the
inflammatory disease process, this finding
supports the notion that a patients volume
of fat tissue could have implications for other
immune-mediated inflammatory conditions
treated with TNF antagonists
34
.
Some of the residual (or unexplained)
variability ascribed to intrinsic and extrinsic
factors might, in reality, be due to genetic or
behavioural factors but is presently not rec-
ognized as such. Hence, it will be impossible
to address these modifiers of benefitrisk,
and eliminating all biological variability of
drug response will remain an elusivegoal.
Behaviour
Prescribing and drug handling in the medi-
cation use system. Regulators approve drugs
for defined indications and conditions, but
the drugs everyday use in the medication
use system is a different matter. We refer to
the medication use system as the complex
collaboration among physicians, nurses,
pharmacists, hospital or nursing home
staff and other health-care professionals
who influence drug utilization in everyday
practice
5
.
A large fraction of all prescriptions are
written off-label, sometimes encouraged
by inappropriate drug company marketing
and promotions efforts
35
. Off-label implies
prescribing for an unapproved condition,
dose regimen, treatment duration, route of
administration or for patients who should
not be given the drug because of existing
contraindications or interacting concomitant
treatment. Some off-label uses may follow
evidence of the drugs efficacy in unapproved
conditions, or may simply happen ahead of
a regulatory label change, and as such may
benefit patients. However, off-label use was
Table 1 | Sources of variability in drug response that may cause toxicity or lack of efficacy
Source of variability Mechanism Drug therapy Description of problem
Biology
Genomics Individual patients genomic make-up
influences PK and PD, affecting drug
concentration profile at the target site and
the likelihood and magnitude of desired and
adverse effects
Trastuzumab PD: only effective in patients overexpressing HER2
receptor on tumour cells
Abacavir PD: High risk of severe hypersensitivity reaction in
patients with HLAB*5701 allele
Codeine PK: lack of analgesia in carriers of nonfunctional
CYP2D6 alleles, toxicity with multiple CYP2D6 gene
copies
Other intrinsic and
extrinsic factors
Comorbidity, baseline severity of disease,
other altered physiological states, or external
factors influencing PK and/or PD
Insulin PD: glucose control and risk of hypoglycaemia
affected by stress or physical activity
Several drugs PK: increased toxicity due to increased absorption
with concomitant consumption of grapefruit juice
Behaviour
Prescribing and drug
handling
Inappropriate or off-label prescribing,
co-prescribing with an interacting drug,
continued prescribing to non-responders or
medication errors
Cerivastatin Rhabdomyolysis due to too high starting doses
Gemfibrozil Interactions in drug label ignored
Mibefradil Toxicity due to drugdrug interactions, label often
ignored
Patient adherence Poor adherence to prescribed drug regimen,
discontinuation (non-persistence), drug
holidays or inadvertent overdosing
Anti-hypertensive
drugs
Non-adherence or non-persistence perceived as
treatment-resistant hypertension
Anti-infective
therapies
Drug holidays leading to the development of
resistance
The overall variability and the extent of the efficacyeffectiveness gap is driven by the interplay between biological and behavioural factors. CYP2D6, cytochrome
P450 2D6; HLA, human leukocyte antigen; PD, pharmacodynamics; PK, pharmacokinetics.
PERSPECTI VES
nrd_3501_jul11.indd 498 20/06/2011 16:34
of health care; for the second question, the
key players will be developers of e-healthcare,
physicians and other health professionals,
patients, regulators and payers of healthcare.
In the following section we will discuss
how the issues framed by these two ques-
tions might successfully be addressed.
Which patient for this drug?
The starting point with this approach is the
knowledge base on a drug, and the goal is
to optimize the treatable patient popula-
tion in an attempt to minimize the number
needed to treat (NNT) and/or to maximize
the number needed to harm (NNH)
66
.
Pharmacogenomics has considerable poten-
tial to stratify patients with a given condition
into treatment-eligible versus non-eligible
groups. The examples of trastuzumab and
abacavir show the power of predictive
pharmacogenomic biomarkers, but experi-
ence with pharmacogenomic (and in some
instances also phenotypic) biomarkers has
also highlighted a number of challenges for
drug regulators.
In the past, many pharmacogenomic bio-
markers have been identified retrospectively
that is, during the course of the develop-
ment of the drug in question and much of
the data presented to regulatory authorities
in support of such biomarkers raised ques-
tions of methodological validity and clinical
utility
19,67
.
The difficulty of ensuring an adequate
predictive value of the biomarker test, also
known as the companion diagnostic test, is
shown by the pitfalls of using immunohisto-
chemistry (IHC), a test with less than perfect
predictive value to guide EGFR-targeted
therapy. Some real target patients will remain
unidentified and will be denied effective
treatment (false-negative patients), whereas
other patients are unnecessarily exposed to
a drug with no benefits but potential adverse
effects (false-positive patients)
19
.
A different challenge is represented by
the question: how big a difference between
the subpopulations identified by a com-
panion diagnostic justifies a label claim?
To illustrate this issue we return to the gefi-
tinib case study. At least one controlled trial
showed that in patients with mutant EGFR,
progression free survival (PFS) was clearly
superior in the gefitinib group compared to
chemotherapy control, but PFS was inferior
to control in mutation-negative patients
68
.
Here, the biomarker provides a clear deci-
sion rule, as it differentiates subpopulations
for which the benefitrisk is dramatically
different. By contrast, an example of a less
than clear-cut genomic biomarker was
reported by Donnelly etal.
69
, who stud-
ied the association of a single nucleotide
polymorphism in the 3-hydroxy-3-methyl-
glutaryl coenzyme A reductase (HMGCR)
gene with the lipid-lowering response to
statins. Retrospective analysis identified
two strata of patients, one of which reached
the total cholesterol target level in 71.9% of
the cases, versus 49.0% in the other group
69
.
Considering the relatively benign safety
profile of statins at appropriate doses, one
may conclude that the benefitrisk is posi-
tive for both populations, and restricting
the label to the higher response stratum
may be unjustified. This case is not unique,
and several potential biomarkers do not
identify clearly positive or negative ben-
efitrisk strata, but merely define different
patient subpopulations with marginally
different benefitrisk. For clinicians and
patients, such biomarkers may have no
added value, but payers of health care
might decide to limit coverage of that drug
to the biomarker-defined high responder
group because all else being equal
cost effectiveness will be superior in that
group. This may even have negative con-
sequences, as biomarker stratification may
come to be seen not as a means to optimize
benefitrisk but to exclude patients from
(expensive) treatment. There is at least
one reported precedent of a payer using a
pharmacogenomic biomarker that is not
on the label of the drug to increase the cost
effectiveness of a therapy
19
. The widely
publicised debate around the clinical added
value of cytochrome P450 2C9 (CYP2C9)
and vitamin K epoxide reductase complex
subunit 1 (VKORC1) genotyping for bet-
ter dosing of the anticoagulant warfarin is
another useful case study demonstrating
the regulators challenges with tests that
address only part of the expected variability
in drug response. (For reviews of the warfa-
rin case, please refer to REFS7072.)
The availability of less than clear-cut pre-
dictive markers raises the question of whether
marker-negative patients can or should be
excluded from pivotal clinical trials. The
conundrum here is that focusing on expected
high responder groups will make the clini-
cal development phase faster and cheaper,
support the drug development endeavour
and bring novel treatment options to (some)
patients earlier; however, it may be ethically
challenging to exclude patients who may
potentially benefit, as in the case of HMGCR
gene stratification describedabove.
Solutions to these issues are, however,
beginning to emerge. Industry and the regu-
latory community have learned from past
experience. The pharmaceutical industry
Table 2 | Restrictions in the wording of regulatory drug labels in 2009
Indication as shown on the label or SmPC No. of drugs in United States
(n=25)
No. of drugs in EU
(n=29)
Broad, unrestricted* (for example, benign prostate hyperplasia) 18 17
Restricted to second-line (for example, carcinoma after failure
of prior platinum-containing regimen)
5 6
Restricted to advanced stage of disease (for example, cartilage
defects of grade III or IV)
2 2
Restricted to combination therapy (for example, rheumatoid
arthritis in combination with methotrexate)
0 3
Restricted to a pharmacogenomic subset (for example,
cancer with activating mutations of EGFR-TK)
0 1
For all new drugs approved in 2009, we reviewed the official labels (for US drugs) or the summary of product characteristics (SmPC) (for European Union (EU) drugs).
(For the EU, we reviewed only drugs that were approved by centralized procedure.) The list of new biologics and new molecular entities approved by the US Food
and Drug Administration (FDA) was obtained from REF.108, and the list of new active substances for the EU was obtained from REF.109. Label/SmPC information
was obtained from the websites of the FDA and European Medicines Agency (EMA). Only wording under the headings of Indications and usage (FDA) or
Therapeutic indication (EMA) was considered. Note that contraindications and warnings, which may also restrict the treatment-eligible population, were not
considered for this analysis. EGFR-TK, epidermal growth factor receptor tyrosine kinase. *Restrictions to age groups are also included under this heading (for
example, not approved for use in pediatric patients).
PERSPECTI VES
nrd_3501_jul11.indd 501 20/06/2011 16:34
effectiveness trials might allow post-reg-
ulatory decision makers, such as payers of
health care or clinicians, to better judge to
whom results can reasonably be applied
to in a realistic health-care setting. However,
that benefit would come at a high price.
To explain the downside of a regulatory
effectiveness paradigm, we recall that inter-
pretation of RCT results can be considered
as a signal-to-noise problem. The majority
of regulatory clinical trials are superiority
showing that is, they aim to demonstrate
a statistically significant positive differ-
ence in observed outcome between a group
of patients receiving the test drug and a
control group, who are most often given
placebo. The higher the variability in the
trial (noise), the less likely it is that a differ-
ence (signal) can be shown even where
it exists. For instance, according to Koops
Law, poor adherence to the experimental
drug regimen by some patients will increase
variability and diminish the average effect
size. Sponsors of clinical trials are very
much aware of that risk and frequently seek
to maximize patient adherence by way of
patient reminders, frequent office visits or
building an initial placebo run-in phase into
the trial protocol; in the run-in phase, all
eligible patients receive placebo, and those
who are poorly compliant are excluded
before randomization. Such measures, often
referred to as enrichment (REF.57), are
expected to increase the signal-to-noise ratio
and the chance of the trial to be successful,
but they reduce external validity. Threats
to the signal-to-noise ratio may also come
from overly broad patient selection criteria,
such as the inadvertent inclusion of patients
with a viral infection into a study of a novel
antibacterialagent.
Effectiveness trials, in which condi-
tions are less stringently controlled, have
higher noise levels and the signal may be
reduced or lost in the noise (FIG.2). At a
minimum, this would result in much larger,
more expensive trials to achieve statistical
significance. In the worst case, it poses the
danger of missing potentially useful sub-
stances: a drug might be deemed ineffective
in a noisy effectiveness trial, even though
it may be beneficial to some patients under
some conditions. Depriving these patients
of valuable treatment options would amount
to a false-negative regulatory decision. The
pitfalls of requesting effectiveness data as
a basis for marketing authorization or of
allowing unlimited variation in premarket-
ing RCTs are also recognized by proponents
of PCTs, some of whom advocate PCTs as
more appropriate vehicles for examining
variation in prescriber or patient adherence
(REF. 18). We support this notion and draw
attention to the fact that the demonstration
of a drugs potential for delivering benefits
is a very different question from assessing
whether that potential is achieved in real
life. We argue that, in a situation in which
the internal validity of pre-licensing RCTs is
not in doubt but where there is an apparent
large gap between efficacy and effectiveness,
one is not looking at a drug problem but
at a health-care delivery problem, and the
focus of remedial action should be shifted to
improving real-life performance.
Bring up reality to the level of expectations.
In an ideal world of evidence-based medi-
cine, the label would exactly reflect the evi-
dence base that is, the treatment scenario
of regulatory clinical trials (step 1) and
drug utilization would conform to the label
(step 2). Alas, both steps cannot be easily
realized.
With regards to the first step, defin-
ing an indication strictly in line with the
clinical trial conditions is not an option,
and wording the label is usually a balancing
act for regulators who need to refrain from
overextrapolating the evidence base while
being mindful of practical constraints faced
by prescribers. For instance, one external
commentator chided regulators for overly
narrow labels:
the FDA approves [cancer drugs]
for indications so narrow (an artifact
of their archaic statistical evaluation
approaches), that most cancer drugs
actually are approved for different uses.
A colon cancer drug isnt approved by
[the] FDA to treat any colon cancer
patient at any time, but rather to treat,
for example, only those patients who
have exhausted all other approved
options, because that is the population
it was statistically tested in for approval.
Also, oncologists dont use cancer drugs
strictly in accordance with the FDAs
narrow approvals because doing so
simply wouldnt work to keep their
patients alive. (REF.58)
We reviewed the wording of the treat-
ment indications in the labels of new active
substances authorized in the year 2009 in
the United States and the EU (centralized
procedure only) (TABLE2). Our findings sup-
port the notion that the label is frequently
restricted to what some may consider an
artificial subset because, in most cases, the
limited indication reflects the clinical trial
scenarios. It is indeed biologically plausible
that some of the more narrowly authorized
drugs might be equally safe and effica-
cious in extended settings, such as first-line
therapy or monotherapy. However, too much
extrapolation in the wording of the label
would violate the principles of evidence-
based drug licensing. Overly broad labels
could rightfully be criticized as further
increasing variability and widening the
efficacyeffectiveness gap when extrapola-
tion turns out to be unjustified
59
. The cur-
rent model of drug licensing leaves limited
room for regulators to address the dilemma
between too wide or too narrow labels. It is
hoped that innovative approaches to licens-
ing such as staggered approval (or progres-
sive authorization) may in future better
align the knowledge of a drug with the label
claim
6062
.
The second step in this process ensur-
ing that drug utilization conforms to the
label is even more difficult to achieve as it
requires behavioural changes from all actors
in the medication use system and from
patients. Past attempts at reducing variability
of drug utilization, which were largely based
on motivational and educational efforts,
were not always successful
63
.
These considerations highlight that both
avenues to address the efficacyeffective-
ness gap are fraught with difficulties and
neither is likely to achieve satisfactory results
on its own. The most useful approach will
be to simultaneously address expectations
andreality.
The problem may also be framed from a
different angle. Drug effects good or bad
are the result of a successful or bad match
between a drug and a given patient. Drugs
may be more or less forgiving (REF.64);
those with a steep doseresponse relation-
ship, narrow therapeutic window and rapid
elimination are often less forgiving. Also,
some patients are more unforgiving than
others; for example, some patients may have
genetic traits that predispose them to an
adverse event or they may be taking multiple
concomitant medications. Poor therapeutic
outcomes result when an unforgiving drug
meets an unforgiving patient
65
, and the key
to bridging the efficacyeffectiveness gap is
to avoid poor matches. We propose that suc-
cessful attempts to this end will have to be
centred on two questions: which patient for
this drug? and which drug regimen for this
patient? Novel technologies will have to be
harnessed both for the pre- and post-licens-
ing lifespan of drugs. The call for action to
address the first question is directed mostly
to developers of drugs, regulators and payers
PERSPECTI VES
nrd_3501_jul11.indd 500 20/06/2011 16:34
of health care; for the second question, the
key players will be developers of e-healthcare,
physicians and other health professionals,
patients, regulators and payers of healthcare.
In the following section we will discuss
how the issues framed by these two ques-
tions might successfully be addressed.
Which patient for this drug?
The starting point with this approach is the
knowledge base on a drug, and the goal is
to optimize the treatable patient popula-
tion in an attempt to minimize the number
needed to treat (NNT) and/or to maximize
the number needed to harm (NNH)
66
.
Pharmacogenomics has considerable poten-
tial to stratify patients with a given condition
into treatment-eligible versus non-eligible
groups. The examples of trastuzumab and
abacavir show the power of predictive
pharmacogenomic biomarkers, but experi-
ence with pharmacogenomic (and in some
instances also phenotypic) biomarkers has
also highlighted a number of challenges for
drug regulators.
In the past, many pharmacogenomic bio-
markers have been identified retrospectively
that is, during the course of the develop-
ment of the drug in question and much of
the data presented to regulatory authorities
in support of such biomarkers raised ques-
tions of methodological validity and clinical
utility
19,67
.
The difficulty of ensuring an adequate
predictive value of the biomarker test, also
known as the companion diagnostic test, is
shown by the pitfalls of using immunohisto-
chemistry (IHC), a test with less than perfect
predictive value to guide EGFR-targeted
therapy. Some real target patients will remain
unidentified and will be denied effective
treatment (false-negative patients), whereas
other patients are unnecessarily exposed to
a drug with no benefits but potential adverse
effects (false-positive patients)
19
.
A different challenge is represented by
the question: how big a difference between
the subpopulations identified by a com-
panion diagnostic justifies a label claim?
To illustrate this issue we return to the gefi-
tinib case study. At least one controlled trial
showed that in patients with mutant EGFR,
progression free survival (PFS) was clearly
superior in the gefitinib group compared to
chemotherapy control, but PFS was inferior
to control in mutation-negative patients
68
.
Here, the biomarker provides a clear deci-
sion rule, as it differentiates subpopulations
for which the benefitrisk is dramatically
different. By contrast, an example of a less
than clear-cut genomic biomarker was
reported by Donnelly etal.
69
, who stud-
ied the association of a single nucleotide
polymorphism in the 3-hydroxy-3-methyl-
glutaryl coenzyme A reductase (HMGCR)
gene with the lipid-lowering response to
statins. Retrospective analysis identified
two strata of patients, one of which reached
the total cholesterol target level in 71.9% of
the cases, versus 49.0% in the other group
69
.
Considering the relatively benign safety
profile of statins at appropriate doses, one
may conclude that the benefitrisk is posi-
tive for both populations, and restricting
the label to the higher response stratum
may be unjustified. This case is not unique,
and several potential biomarkers do not
identify clearly positive or negative ben-
efitrisk strata, but merely define different
patient subpopulations with marginally
different benefitrisk. For clinicians and
patients, such biomarkers may have no
added value, but payers of health care
might decide to limit coverage of that drug
to the biomarker-defined high responder
group because all else being equal
cost effectiveness will be superior in that
group. This may even have negative con-
sequences, as biomarker stratification may
come to be seen not as a means to optimize
benefitrisk but to exclude patients from
(expensive) treatment. There is at least
one reported precedent of a payer using a
pharmacogenomic biomarker that is not
on the label of the drug to increase the cost
effectiveness of a therapy
19
. The widely
publicised debate around the clinical added
value of cytochrome P450 2C9 (CYP2C9)
and vitamin K epoxide reductase complex
subunit 1 (VKORC1) genotyping for bet-
ter dosing of the anticoagulant warfarin is
another useful case study demonstrating
the regulators challenges with tests that
address only part of the expected variability
in drug response. (For reviews of the warfa-
rin case, please refer to REFS7072.)
The availability of less than clear-cut pre-
dictive markers raises the question of whether
marker-negative patients can or should be
excluded from pivotal clinical trials. The
conundrum here is that focusing on expected
high responder groups will make the clini-
cal development phase faster and cheaper,
support the drug development endeavour
and bring novel treatment options to (some)
patients earlier; however, it may be ethically
challenging to exclude patients who may
potentially benefit, as in the case of HMGCR
gene stratification describedabove.
Solutions to these issues are, however,
beginning to emerge. Industry and the regu-
latory community have learned from past
experience. The pharmaceutical industry
Table 2 | Restrictions in the wording of regulatory drug labels in 2009
Indication as shown on the label or SmPC No. of drugs in United States
(n=25)
No. of drugs in EU
(n=29)
Broad, unrestricted* (for example, benign prostate hyperplasia) 18 17
Restricted to second-line (for example, carcinoma after failure
of prior platinum-containing regimen)
5 6
Restricted to advanced stage of disease (for example, cartilage
defects of grade III or IV)
2 2
Restricted to combination therapy (for example, rheumatoid
arthritis in combination with methotrexate)
0 3
Restricted to a pharmacogenomic subset (for example,
cancer with activating mutations of EGFR-TK)
0 1
For all new drugs approved in 2009, we reviewed the official labels (for US drugs) or the summary of product characteristics (SmPC) (for European Union (EU) drugs).
(For the EU, we reviewed only drugs that were approved by centralized procedure.) The list of new biologics and new molecular entities approved by the US Food
and Drug Administration (FDA) was obtained from REF.108, and the list of new active substances for the EU was obtained from REF.109. Label/SmPC information
was obtained from the websites of the FDA and European Medicines Agency (EMA). Only wording under the headings of Indications and usage (FDA) or
Therapeutic indication (EMA) was considered. Note that contraindications and warnings, which may also restrict the treatment-eligible population, were not
considered for this analysis. EGFR-TK, epidermal growth factor receptor tyrosine kinase. *Restrictions to age groups are also included under this heading (for
example, not approved for use in pediatric patients).
PERSPECTI VES
nrd_3501_jul11.indd 501 20/06/2011 16:34
effectiveness trials might allow post-reg-
ulatory decision makers, such as payers of
health care or clinicians, to better judge to
whom results can reasonably be applied
to in a realistic health-care setting. However,
that benefit would come at a high price.
To explain the downside of a regulatory
effectiveness paradigm, we recall that inter-
pretation of RCT results can be considered
as a signal-to-noise problem. The majority
of regulatory clinical trials are superiority
showing that is, they aim to demonstrate
a statistically significant positive differ-
ence in observed outcome between a group
of patients receiving the test drug and a
control group, who are most often given
placebo. The higher the variability in the
trial (noise), the less likely it is that a differ-
ence (signal) can be shown even where
it exists. For instance, according to Koops
Law, poor adherence to the experimental
drug regimen by some patients will increase
variability and diminish the average effect
size. Sponsors of clinical trials are very
much aware of that risk and frequently seek
to maximize patient adherence by way of
patient reminders, frequent office visits or
building an initial placebo run-in phase into
the trial protocol; in the run-in phase, all
eligible patients receive placebo, and those
who are poorly compliant are excluded
before randomization. Such measures, often
referred to as enrichment (REF.57), are
expected to increase the signal-to-noise ratio
and the chance of the trial to be successful,
but they reduce external validity. Threats
to the signal-to-noise ratio may also come
from overly broad patient selection criteria,
such as the inadvertent inclusion of patients
with a viral infection into a study of a novel
antibacterialagent.
Effectiveness trials, in which condi-
tions are less stringently controlled, have
higher noise levels and the signal may be
reduced or lost in the noise (FIG.2). At a
minimum, this would result in much larger,
more expensive trials to achieve statistical
significance. In the worst case, it poses the
danger of missing potentially useful sub-
stances: a drug might be deemed ineffective
in a noisy effectiveness trial, even though
it may be beneficial to some patients under
some conditions. Depriving these patients
of valuable treatment options would amount
to a false-negative regulatory decision. The
pitfalls of requesting effectiveness data as
a basis for marketing authorization or of
allowing unlimited variation in premarket-
ing RCTs are also recognized by proponents
of PCTs, some of whom advocate PCTs as
more appropriate vehicles for examining
variation in prescriber or patient adherence
(REF. 18). We support this notion and draw
attention to the fact that the demonstration
of a drugs potential for delivering benefits
is a very different question from assessing
whether that potential is achieved in real
life. We argue that, in a situation in which
the internal validity of pre-licensing RCTs is
not in doubt but where there is an apparent
large gap between efficacy and effectiveness,
one is not looking at a drug problem but
at a health-care delivery problem, and the
focus of remedial action should be shifted to
improving real-life performance.
Bring up reality to the level of expectations.
In an ideal world of evidence-based medi-
cine, the label would exactly reflect the evi-
dence base that is, the treatment scenario
of regulatory clinical trials (step 1) and
drug utilization would conform to the label
(step 2). Alas, both steps cannot be easily
realized.
With regards to the first step, defin-
ing an indication strictly in line with the
clinical trial conditions is not an option,
and wording the label is usually a balancing
act for regulators who need to refrain from
overextrapolating the evidence base while
being mindful of practical constraints faced
by prescribers. For instance, one external
commentator chided regulators for overly
narrow labels:
the FDA approves [cancer drugs]
for indications so narrow (an artifact
of their archaic statistical evaluation
approaches), that most cancer drugs
actually are approved for different uses.
A colon cancer drug isnt approved by
[the] FDA to treat any colon cancer
patient at any time, but rather to treat,
for example, only those patients who
have exhausted all other approved
options, because that is the population
it was statistically tested in for approval.
Also, oncologists dont use cancer drugs
strictly in accordance with the FDAs
narrow approvals because doing so
simply wouldnt work to keep their
patients alive. (REF.58)
We reviewed the wording of the treat-
ment indications in the labels of new active
substances authorized in the year 2009 in
the United States and the EU (centralized
procedure only) (TABLE2). Our findings sup-
port the notion that the label is frequently
restricted to what some may consider an
artificial subset because, in most cases, the
limited indication reflects the clinical trial
scenarios. It is indeed biologically plausible
that some of the more narrowly authorized
drugs might be equally safe and effica-
cious in extended settings, such as first-line
therapy or monotherapy. However, too much
extrapolation in the wording of the label
would violate the principles of evidence-
based drug licensing. Overly broad labels
could rightfully be criticized as further
increasing variability and widening the
efficacyeffectiveness gap when extrapola-
tion turns out to be unjustified
59
. The cur-
rent model of drug licensing leaves limited
room for regulators to address the dilemma
between too wide or too narrow labels. It is
hoped that innovative approaches to licens-
ing such as staggered approval (or progres-
sive authorization) may in future better
align the knowledge of a drug with the label
claim
6062
.
The second step in this process ensur-
ing that drug utilization conforms to the
label is even more difficult to achieve as it
requires behavioural changes from all actors
in the medication use system and from
patients. Past attempts at reducing variability
of drug utilization, which were largely based
on motivational and educational efforts,
were not always successful
63
.
These considerations highlight that both
avenues to address the efficacyeffective-
ness gap are fraught with difficulties and
neither is likely to achieve satisfactory results
on its own. The most useful approach will
be to simultaneously address expectations
andreality.
The problem may also be framed from a
different angle. Drug effects good or bad
are the result of a successful or bad match
between a drug and a given patient. Drugs
may be more or less forgiving (REF.64);
those with a steep doseresponse relation-
ship, narrow therapeutic window and rapid
elimination are often less forgiving. Also,
some patients are more unforgiving than
others; for example, some patients may have
genetic traits that predispose them to an
adverse event or they may be taking multiple
concomitant medications. Poor therapeutic
outcomes result when an unforgiving drug
meets an unforgiving patient
65
, and the key
to bridging the efficacyeffectiveness gap is
to avoid poor matches. We propose that suc-
cessful attempts to this end will have to be
centred on two questions: which patient for
this drug? and which drug regimen for this
patient? Novel technologies will have to be
harnessed both for the pre- and post-licens-
ing lifespan of drugs. The call for action to
address the first question is directed mostly
to developers of drugs, regulators and payers
PERSPECTI VES
nrd_3501_jul11.indd 500 20/06/2011 16:34
Regulators could make a further contribu-
tion to addressing the efficacyeffectiveness
gap: in select cases in which external trial
validity is perceived to be low, or in which
problems associated with everyday clini-
cal use may be anticipated, the assessment
report could provide structured information
on expected relevance of pivotal trials to use
in practice along the lines of, for example,
the PRECIS framework
56
. At least one post-
licensing decision maker, the French Health
Technology Assessment body HAS (French
National Authority for Health), which advises
on the effectiveness and relative effectiveness
of drugs to inform reimbursement and pric-
ing decisions, has proposed a formal meth-
odological framework to predict effectiveness
based on regulatory efficacy trials and expert
input
81
. Time will tell whether this or similar
methods could be more widely used at the
time of licensing.
Which drug regimen for this patient?
The right drug, at the right dose, at the
right time, to the right patient has been at
the core of clinical pharmacology teaching
for decades
82
; it implies the consideration
of as many genotypic, phenotypic and envi-
ronmental patient variables as possible to
individualize drug regimes, considering all
available treatmentoptions.
Industry and regulators aim to design
drug development plans to generate the
information required for tailored dose
regimens to the extent that this is
possible. Covariate-adjusted analysis of
clinical trials has a long tradition in drug
development, and is often used to identify
patients for whom a treatment is more or
less beneficial
83
. Considerable progress
has been made over the past decade or
two in the use of pharmacokinetics/phar-
macodynamics modelling and simulation
(M&S) in drug development to study the
interrelationships among multiple factors
such as gender, age, feeding and fasting,
and comorbidities
84
. The use of M&S has
been encouraged by regulators for optimiz-
ing drug development and recommending
dose adjustments in defined subpopula-
tions, such as children, without specific,
full-powered studies
85
. It will be interesting
to explore how such methodologies could
be applied to observational post-marketing
studies aswell.
However, regulators, industry and payers
of health care can make further contribu-
tions, beyond intelligent pre-licensing trial
designs and analyses, to addressing the
behavioural side of the efficacyeffectiveness
gap (TABLE1).
Quality of prescribing. There has been no
shortage of attempts to promote evidence-
based prescribing and reduce inconsistency
in everyday patient care; training pro-
grammes, dedicated prescriber communi-
cations from regulators and industry, and
clinical practice guidelines have increasingly
been developed
8688
. However, many of these
efforts have had a limited effect on chang-
ing prescriber behaviour, and guideline
adoption is variable
89,90
, but the impact can
be substantial if appropriate measures are
taken, including a focus on ownership
91
. The
reasons for failure are multiple but include
the increasingly demanding health-care
environment for prescribers because of the
widening choice of medicines available,
expanding indications for drug treatment,
greater complexity of treatment regimens
and the pace of change in therapeutics; new
evidence on efficacy or emerging safety
signals mean that what is considered good
prescribing today may not necessarily be
considered so in a years time. Providing
health-care professionals with only printed
reference information, such as formularies
or Dear Health-Care Professional commu-
nications, may no longer be fit for purpose.
The information will not be available when
needed; prescribers simply do not have the
time to consult it as they make rapid point
of care decisions
63
. Against this background,
the burgeoning field of e-healthcare offers
unique opportunities to address prescribing
and medication errors. Advances with the
development of electronic health records,
computerized physician order entry (e-pre-
scribing), and decision support systems pro-
vide opportunities of linking patient-specific
information (for example, comorbidity and
renal insufficiency) with electronic drug-
specific information (for example, contrain-
dications and dose-modification guidance
92
).
Bringing these building blocks of e-health-
care together can help prescribers to improve
drug regimens by including patient-specific
alerts, such as for contraindications, limits of
recommended treatment duration or poten-
tial interactions between drugs prescribed by
different specialists (FIG.4).
Pilot schemes of complete computerized
physician order entry and e-prescribing
tools in outpatient care are currently being
tested
93,94
. Furthermore, integrated but lim-
ited clinical computerized decision support
applications aimed at guiding clinicians to
appropriately handle drug dosing and drug
interactions are now up and running
95,96
.
However, the full statutory label information
from drug regulatory bodies is not available
in a logical format that can interface with
the electronic health record and decision
support systems. The European Medicines
Agency is exploring ways to make label
information available at the time of licensing
in an electronic format that can be integrated
with the other building blocks of e-health-
care at point ofcare
63
.
Regulators have traditionally confined
themselves to providing information on the
rational use of drugs interference with
everyday prescribing was not part of their
role. Boundaries have somewhat changed
with the establishment of RMPs (in the EU)
and REMSs (in the United States). These
post-marketing activities are not limited to
risk assessment but may include a pro-active
risk mitigation component.
A useful example is the anti-epileptic
agent vigabatrin. Because of the potential
for visual loss, it was approved in the United
States with an REMS. One component of an
REMS is elements to assure safe use, which
function as a type of restricted distribution
system. The particular elements in this case
included special certification of health-care
providers who prescribe vigabatrin, special
certification of pharmacies that dispense
vigabatrin, a requirement that vigabatrin be
dispensed to patients with evidence or other
documentation of safe-use conditions, and
enrolment of each patient using vigabatrin
in a registry. In addition to this, the drug has
a dedicated implementation plan to facili-
tate ongoing benefitrisk assessments; all
patients receive a baseline ophthalmologic
evaluation and regular vision monitoring to
ensure that vigabatrin-induced vision loss
is detected as early as possible, and vigaba-
trin therapy is discontinued in patients
who experience an inadequate clinical
response
5
. This relatively new approach to
risk mitigation initiated by regulators may
be interpreted as interference with everyday
practice, but it is hoped that it will contribute
to a favourable benefitrisk profile in real
life, reducing the efficacyeffectivenessgap.
Patient adherence. The impact of poor
adherence on public health and the cost to
the health-care system are well understood
97
.
However, considerations of adherence do
not figure prominently in the evaluation
of pre-licensing trials, and the regulatory
analytical standard, the intention-to-treat
(ITT) analysis, is deliberately blind as to
whether individual patients followed the
treatment protocol. For some treatment
scenarios (such as time consuming, mul-
tiple drug inhalations for children and
adolescents with cystic fibrosis)
98
, it may be
justified to make trials adherence-informed,
PERSPECTI VES
nrd_3501_jul11.indd 503 20/06/2011 16:34
lniiul, nurrov liconso
(suqo 1)
Widor uion
ouluion
(lor oxumlo,
includinq
murlor-
noquivo
uions )
Clinicul
riuls undor
os-murloinq
commimon
lollov-on, vidor
liconso(s) (suqo 2,
und so on)
Lnricbod uion
ouluion vib bos
oxocod bono-risl
(lor oxumlo,
murlor-osiivo uions)
approach to stratified medicine has changed
fundamentally. In the past, retrospective
identification of responder populations was
sometimes a last resort for drugs that might
have failed for broader populations, but
now many research-based pharmaceutical
companies have modified their approach to
drug development and will routinely pursue
pro-active diagnostic biomarker strategies,
including DNA collection in all trials
19,73
.
This evolution may also reflect a more
fundamental realization by the drug devel-
opment community that the blockbuster
paradigm of old the attempt to maximize
the treatment-eligible population may
not be a winning strategy, and that the goal
should be to optimize the treatment-eligible
population. The change of mindsets is sup-
ported by increasingly powerful tools for
high-throughput genotyping, allowing drug
developers to economically assess ever more
genomic markers
74
.
Progress is also being made in the field
of co-development of targeted drugs and
their companion diagnostics. Regulators
have provided reflections on the key sci-
entific principles that need to be met to
ensure that the performance of the chosen
assay is sufficiently reliable to optimize drug
development and regulatory submission
67,75
.
Moreover, agencies have established consul-
tation procedures to enable the development
and qualification of biomarkers
76,77
.
Some adaptation of the current drug
regulatory paradigm may also be called for
to achieve the competing objectives of: full
exploitation of biomarker strategies for opti-
mization of the treatment-eligible popula-
tion; balancing the need for early access by
patients with a high degree of unmet need
with the ethical imperative to not exclude
other patients from potentially beneficial
treatments; and keeping the drug develop-
ment process efficient and sustainable.
An expert panel has recently proposed a
strategy for the approval of targeted cancer
drugs
73,78
. We would broaden that strategy
to include drugs for non-cancer conditions
and outline what may be called the onion
skin model of drug licensing (FIG.3). In a
drug development programme with a rea-
sonably good a priori chance of identifying
a high benefitrisk subgroup, initial drug
development would focus on marker posi-
tive patients. Marker-negative patients need
not be included in the initial study phase.
It follows that, in the first stage, the marker
need not be shown to predict lack of effect
in the marker-negative population. If the
benefitrisk profile was deemed favour-
able in marker-positive patients, an initial,
narrow license would be granted. In the next
stage, the company would need to assess, as
a post-marketing study, benefitrisk in the
marker-negative stratum (or define the use-
fulness of the test to predict lack of efficacy).
If successful, the label will be broadened to
include a larger treatment-eligible popula-
tion. The second stage is required only if
ethically acceptable, that is, in the absence of
convincing arguments that patients may be
unduly harmed by enrolment in these post-
marketing studies. In some instances, the
goal might most efficiently be achieved by
way of adaptive clinical trial design
78
.
The concept of staggered approval that
is, prioritizing development to an enriched
population of expected high-responders/
high-toleraters (FIG.3) would mark a
departure from the current practice, in some
therapeutic areas, of initially studying last-
line conditions such as cancer patients who
have failed standard treatment. We acknowl-
edge the ethical challenges, but submit that
the last-line first paradigm of enrolling
patients with multiresistant (for example,
cancer) or advanced (for example, autoim-
mune or degenerative) pathologies may
destine to fail potentially useful first-line
treatments
79,80
. Most likely, this form of stag-
gered approval will be based on pharmaco-
genomic biomarkers, but it may apply
equally well to phenotypicmarkers.
Biomarker-guided staggered approval
has the potential to reduce variability in
drug response and help to address the effi-
cacyeffectiveness gap, but there is a sine qua
non; drug utilization will have to mirror the
regulatory pathway. In the first instance, this
implies that regulatory and reimbursement
policies will need to be aligned as closely
as possible. Regulators allow access to the
market, but payers control which patient
groups will be able to receive the drug.
Hence, the success of novel development
paradigms will require an ongoing dialogue
between the regulatory and payers commu-
nities in a given jurisdiction. In the second
instance, alignment of the other actors in the
medication use system will be required, as
discussedbelow.
Figure 3 | The onion skin model of drug licensing. This approach to drug development and
licensing focuses initially on the patient strata with the best expected benefitrisk profile instead of
the sometimes practiced last-line first strategy. The gains from this initially limited focus are smaller
clinical trials and faster access for those patients who are expected to benefit most. The enriched
patient sample is selected by way of a (genotypic or phenotypic) biomarker. Biomarker-negative
patients need not be included in the initial study phase. If the benefitrisk profile was deemed
favourable in marker-positive patients, an initial, narrow license would be granted. In the next stage,
the company would need to assess, as a post-marketing commitment, benefitrisk in the marker-
negative strata. If successful, the label will be broadened to include a larger treatment-eligible
population. A drawback of this regulatory pathway is that the positive and negative predictive values
of the biomarker cannot be estimated during the initial phases of the development and licensing
process; the approach would most likely be justified and useful for drugs that have the potential to
address a high level of unmet medical need.
PERSPECTI VES
nrd_3501_jul11.indd 502 20/06/2011 16:34
Regulators could make a further contribu-
tion to addressing the efficacyeffectiveness
gap: in select cases in which external trial
validity is perceived to be low, or in which
problems associated with everyday clini-
cal use may be anticipated, the assessment
report could provide structured information
on expected relevance of pivotal trials to use
in practice along the lines of, for example,
the PRECIS framework
56
. At least one post-
licensing decision maker, the French Health
Technology Assessment body HAS (French
National Authority for Health), which advises
on the effectiveness and relative effectiveness
of drugs to inform reimbursement and pric-
ing decisions, has proposed a formal meth-
odological framework to predict effectiveness
based on regulatory efficacy trials and expert
input
81
. Time will tell whether this or similar
methods could be more widely used at the
time of licensing.
Which drug regimen for this patient?
The right drug, at the right dose, at the
right time, to the right patient has been at
the core of clinical pharmacology teaching
for decades
82
; it implies the consideration
of as many genotypic, phenotypic and envi-
ronmental patient variables as possible to
individualize drug regimes, considering all
available treatmentoptions.
Industry and regulators aim to design
drug development plans to generate the
information required for tailored dose
regimens to the extent that this is
possible. Covariate-adjusted analysis of
clinical trials has a long tradition in drug
development, and is often used to identify
patients for whom a treatment is more or
less beneficial
83
. Considerable progress
has been made over the past decade or
two in the use of pharmacokinetics/phar-
macodynamics modelling and simulation
(M&S) in drug development to study the
interrelationships among multiple factors
such as gender, age, feeding and fasting,
and comorbidities
84
. The use of M&S has
been encouraged by regulators for optimiz-
ing drug development and recommending
dose adjustments in defined subpopula-
tions, such as children, without specific,
full-powered studies
85
. It will be interesting
to explore how such methodologies could
be applied to observational post-marketing
studies aswell.
However, regulators, industry and payers
of health care can make further contribu-
tions, beyond intelligent pre-licensing trial
designs and analyses, to addressing the
behavioural side of the efficacyeffectiveness
gap (TABLE1).
Quality of prescribing. There has been no
shortage of attempts to promote evidence-
based prescribing and reduce inconsistency
in everyday patient care; training pro-
grammes, dedicated prescriber communi-
cations from regulators and industry, and
clinical practice guidelines have increasingly
been developed
8688
. However, many of these
efforts have had a limited effect on chang-
ing prescriber behaviour, and guideline
adoption is variable
89,90
, but the impact can
be substantial if appropriate measures are
taken, including a focus on ownership
91
. The
reasons for failure are multiple but include
the increasingly demanding health-care
environment for prescribers because of the
widening choice of medicines available,
expanding indications for drug treatment,
greater complexity of treatment regimens
and the pace of change in therapeutics; new
evidence on efficacy or emerging safety
signals mean that what is considered good
prescribing today may not necessarily be
considered so in a years time. Providing
health-care professionals with only printed
reference information, such as formularies
or Dear Health-Care Professional commu-
nications, may no longer be fit for purpose.
The information will not be available when
needed; prescribers simply do not have the
time to consult it as they make rapid point
of care decisions
63
. Against this background,
the burgeoning field of e-healthcare offers
unique opportunities to address prescribing
and medication errors. Advances with the
development of electronic health records,
computerized physician order entry (e-pre-
scribing), and decision support systems pro-
vide opportunities of linking patient-specific
information (for example, comorbidity and
renal insufficiency) with electronic drug-
specific information (for example, contrain-
dications and dose-modification guidance
92
).
Bringing these building blocks of e-health-
care together can help prescribers to improve
drug regimens by including patient-specific
alerts, such as for contraindications, limits of
recommended treatment duration or poten-
tial interactions between drugs prescribed by
different specialists (FIG.4).
Pilot schemes of complete computerized
physician order entry and e-prescribing
tools in outpatient care are currently being
tested
93,94
. Furthermore, integrated but lim-
ited clinical computerized decision support
applications aimed at guiding clinicians to
appropriately handle drug dosing and drug
interactions are now up and running
95,96
.
However, the full statutory label information
from drug regulatory bodies is not available
in a logical format that can interface with
the electronic health record and decision
support systems. The European Medicines
Agency is exploring ways to make label
information available at the time of licensing
in an electronic format that can be integrated
with the other building blocks of e-health-
care at point ofcare
63
.
Regulators have traditionally confined
themselves to providing information on the
rational use of drugs interference with
everyday prescribing was not part of their
role. Boundaries have somewhat changed
with the establishment of RMPs (in the EU)
and REMSs (in the United States). These
post-marketing activities are not limited to
risk assessment but may include a pro-active
risk mitigation component.
A useful example is the anti-epileptic
agent vigabatrin. Because of the potential
for visual loss, it was approved in the United
States with an REMS. One component of an
REMS is elements to assure safe use, which
function as a type of restricted distribution
system. The particular elements in this case
included special certification of health-care
providers who prescribe vigabatrin, special
certification of pharmacies that dispense
vigabatrin, a requirement that vigabatrin be
dispensed to patients with evidence or other
documentation of safe-use conditions, and
enrolment of each patient using vigabatrin
in a registry. In addition to this, the drug has
a dedicated implementation plan to facili-
tate ongoing benefitrisk assessments; all
patients receive a baseline ophthalmologic
evaluation and regular vision monitoring to
ensure that vigabatrin-induced vision loss
is detected as early as possible, and vigaba-
trin therapy is discontinued in patients
who experience an inadequate clinical
response
5
. This relatively new approach to
risk mitigation initiated by regulators may
be interpreted as interference with everyday
practice, but it is hoped that it will contribute
to a favourable benefitrisk profile in real
life, reducing the efficacyeffectivenessgap.
Patient adherence. The impact of poor
adherence on public health and the cost to
the health-care system are well understood
97
.
However, considerations of adherence do
not figure prominently in the evaluation
of pre-licensing trials, and the regulatory
analytical standard, the intention-to-treat
(ITT) analysis, is deliberately blind as to
whether individual patients followed the
treatment protocol. For some treatment
scenarios (such as time consuming, mul-
tiple drug inhalations for children and
adolescents with cystic fibrosis)
98
, it may be
justified to make trials adherence-informed,
PERSPECTI VES
nrd_3501_jul11.indd 503 20/06/2011 16:34
lniiul, nurrov liconso
(suqo 1)
Widor uion
ouluion
(lor oxumlo,
includinq
murlor-
noquivo
uions )
Clinicul
riuls undor
os-murloinq
commimon
lollov-on, vidor
liconso(s) (suqo 2,
und so on)
Lnricbod uion
ouluion vib bos
oxocod bono-risl
(lor oxumlo,
murlor-osiivo uions)
approach to stratified medicine has changed
fundamentally. In the past, retrospective
identification of responder populations was
sometimes a last resort for drugs that might
have failed for broader populations, but
now many research-based pharmaceutical
companies have modified their approach to
drug development and will routinely pursue
pro-active diagnostic biomarker strategies,
including DNA collection in all trials
19,73
.
This evolution may also reflect a more
fundamental realization by the drug devel-
opment community that the blockbuster
paradigm of old the attempt to maximize
the treatment-eligible population may
not be a winning strategy, and that the goal
should be to optimize the treatment-eligible
population. The change of mindsets is sup-
ported by increasingly powerful tools for
high-throughput genotyping, allowing drug
developers to economically assess ever more
genomic markers
74
.
Progress is also being made in the field
of co-development of targeted drugs and
their companion diagnostics. Regulators
have provided reflections on the key sci-
entific principles that need to be met to
ensure that the performance of the chosen
assay is sufficiently reliable to optimize drug
development and regulatory submission
67,75
.
Moreover, agencies have established consul-
tation procedures to enable the development
and qualification of biomarkers
76,77
.
Some adaptation of the current drug
regulatory paradigm may also be called for
to achieve the competing objectives of: full
exploitation of biomarker strategies for opti-
mization of the treatment-eligible popula-
tion; balancing the need for early access by
patients with a high degree of unmet need
with the ethical imperative to not exclude
other patients from potentially beneficial
treatments; and keeping the drug develop-
ment process efficient and sustainable.
An expert panel has recently proposed a
strategy for the approval of targeted cancer
drugs
73,78
. We would broaden that strategy
to include drugs for non-cancer conditions
and outline what may be called the onion
skin model of drug licensing (FIG.3). In a
drug development programme with a rea-
sonably good a priori chance of identifying
a high benefitrisk subgroup, initial drug
development would focus on marker posi-
tive patients. Marker-negative patients need
not be included in the initial study phase.
It follows that, in the first stage, the marker
need not be shown to predict lack of effect
in the marker-negative population. If the
benefitrisk profile was deemed favour-
able in marker-positive patients, an initial,
narrow license would be granted. In the next
stage, the company would need to assess, as
a post-marketing study, benefitrisk in the
marker-negative stratum (or define the use-
fulness of the test to predict lack of efficacy).
If successful, the label will be broadened to
include a larger treatment-eligible popula-
tion. The second stage is required only if
ethically acceptable, that is, in the absence of
convincing arguments that patients may be
unduly harmed by enrolment in these post-
marketing studies. In some instances, the
goal might most efficiently be achieved by
way of adaptive clinical trial design
78
.
The concept of staggered approval that
is, prioritizing development to an enriched
population of expected high-responders/
high-toleraters (FIG.3) would mark a
departure from the current practice, in some
therapeutic areas, of initially studying last-
line conditions such as cancer patients who
have failed standard treatment. We acknowl-
edge the ethical challenges, but submit that
the last-line first paradigm of enrolling
patients with multiresistant (for example,
cancer) or advanced (for example, autoim-
mune or degenerative) pathologies may
destine to fail potentially useful first-line
treatments
79,80
. Most likely, this form of stag-
gered approval will be based on pharmaco-
genomic biomarkers, but it may apply
equally well to phenotypicmarkers.
Biomarker-guided staggered approval
has the potential to reduce variability in
drug response and help to address the effi-
cacyeffectiveness gap, but there is a sine qua
non; drug utilization will have to mirror the
regulatory pathway. In the first instance, this
implies that regulatory and reimbursement
policies will need to be aligned as closely
as possible. Regulators allow access to the
market, but payers control which patient
groups will be able to receive the drug.
Hence, the success of novel development
paradigms will require an ongoing dialogue
between the regulatory and payers commu-
nities in a given jurisdiction. In the second
instance, alignment of the other actors in the
medication use system will be required, as
discussedbelow.
Figure 3 | The onion skin model of drug licensing. This approach to drug development and
licensing focuses initially on the patient strata with the best expected benefitrisk profile instead of
the sometimes practiced last-line first strategy. The gains from this initially limited focus are smaller
clinical trials and faster access for those patients who are expected to benefit most. The enriched
patient sample is selected by way of a (genotypic or phenotypic) biomarker. Biomarker-negative
patients need not be included in the initial study phase. If the benefitrisk profile was deemed
favourable in marker-positive patients, an initial, narrow license would be granted. In the next stage,
the company would need to assess, as a post-marketing commitment, benefitrisk in the marker-
negative strata. If successful, the label will be broadened to include a larger treatment-eligible
population. A drawback of this regulatory pathway is that the positive and negative predictive values
of the biomarker cannot be estimated during the initial phases of the development and licensing
process; the approach would most likely be justified and useful for drugs that have the potential to
address a high level of unmet medical need.
PERSPECTI VES
nrd_3501_jul11.indd 502 20/06/2011 16:34
have discussed the nonlinear interaction
of different sources of variability. It follows
that even large reductions in variance from
a single source may have counter-intuitively
small effects in reducing the overall variance
in drug response
46
. A more holistic effort
will therefore be required from all players
to tackle the efficacyeffectivenessgap.
Considering the time and cost of bringing
new drugs to market, it seems that attempts to
improve drug effectiveness in real life repre-
sent good value for money; this is a space with
low-hanging fruits that might substantially
improve public health. Revisiting Sir William
Oslers observation, when this potential is
realized, medicine might [indeed become]
a science and not an art.
Hans-Georg Eichler, Eric Abadie, Bruno Flamion,
Hubert Leufkens, Christian K.Schneider and Brigitte
Bloechl-Daum are at the European Medicines Agency,
London, UK.
Hans-Georg Eichler is also at the Massachusetts Institute
of Technology, Cambridge, Massachusetts, USA.
Eric Abadie is also at the European Medicines Agency
Committee for Medicinal Products for Human Use,
London, UK; and theGeneral Directorate Agence
Francaise de Securite Sanitaire des Produits de Sant
(AFSSAPS), Paris, France.
Alasdair Breckenridge is at the Medicines and
Healthcare Products Regulatory Agency, London, UK.
Bruno Flamion is also at the Agence Fdrale des
Mdicaments et des Produits de Sant, Brussels,
Belgium.
Lars L.Gustafsson is at the Division of Clinical
Pharmacology, Department of Laboratory Medicine,
Karolinska Institutet, Stockholm, Sweden.
Hubert Leufkens is also at the Utrecht Institute for
Pharmaceutical Sciences (UIPS), Utrecht,
The Netherlands.
Malcolm Rowland is at the Centre for Applied
Pharmacokinetic Research, School of Pharmacy and
Pharmaceutical Sciences, University of Manchester,
Manchester, UK.
Christian K.Schneider is also at the Paul-Ehrlich-Institut,
Langen, Germany; and the Twincore Centre for
Experimental and Clinical Infection Research,
Hannover, Germany.
Brigitte Bloechl-Daum is also at the Department of
Clinical Pharmacology, Medical University Vienna,
Vienna, Austria.
Correspondence to B.B.-D.
e-mail: brigitte.bloechl-daum@meduniwien.ac.at
doi: 10.1038/nrd3501
Disclaimer: The views expressed in this article are the
personal views of the authors and may not be
understood or quoted as being made on behalf of or
reflecting the position of the regulatory agencies,
health technology assessment bodies or other
organizations that the authors work for.
1. Danish Medicines Agency. Conclusions and
recommendations from the Pharmaceutical Forum.
Danish Medicines Agency [online], http://www.dkma.
dk/1024/visUKLSArtikel.asp?artikelID=17481
(2010).
2. Luce, B. R. et al. EBM, HTA, and CER: clearing the
confusion. Millbank Q. 88, 256276 (2010).
3. European Medicines Agency. European Medicines
Agency recommends measures to manage
contamination of heparin-containing medicines.
European Medicines Agency [online], http://www.ema.
europa.eu/ema/index.jsp?curl=pages/news_and_
events/news/2009/11/news_detail_000315.
jsp&murl=menus/news_and_events/news_and_
events.jsp&mid=WC0b01ac058004d5c1&jsenabled
=true (2008).
4. European Medicines Agency. Studies assessed by the
EMEA indicate no increased risk of developing cancer
for patients who have taken Viracept contaminated
with ethyl mesilate. European Medicines Agency
[online]. http://www.ema.europa.eu/ema/index.
jsp?curl=pages/news_and_events/news/2009/11/
news_detail_000292.jsp&murl=menus/news_and_
events/news_and_events.
jsp&mid=WC0b01ac058004d5c1 (2008).
5. Dal Pan, G. J., Blackburn, S. & Karwoski, C. in
Textbook of Pharmacoepidemiology (eds Strom, B. L.
& Kimmel, S. E.) (John Wiley & Sons, New Jersey), (in
the press).
6. Pocock, S. J. & Lubsen, J. More on subgroup analyses
in clinical trials. N.Engl. J.Med. 358, 2076 (2008).
7. Ingelman-Sundberg, M. Pharmacogenetics: an
opportunity for a safer and more efficient
pharmacotherapy. J. Intern. Med. 250, 186200
(2001).
8. Roses, A. D. Pharmacogenetics in drug discovery and
development: a translational perspective. Nature Rev.
Drug Discov. 7, 807817 (2008).
9. Pharoa, P. D. & Hollingworth, W. Cost effectiveness of
lowering cholesterol concentration with statins in
patients with and without pre-existing coronary heart
disease: life table method applied to health authority
population. BMJ 312, 14431448 (1996).
10. Poole, S. G. & Dooley, M. J. Off-label prescribing in
oncology. Support Care Cancer 12, 302305 (2004).
11. Hsien, L. et al. Off-label drug use among hospitalised
children: identifying areas with the highest need for
research. Pharm. World Sci. 30, 497502 (2008).
12. [No authors listed.] Guidance for off-label use of drugs.
Lancet Neurol. 7, 285 (2008).
13. Schosser, R. Risk/benefit evaluation of drugs: the role
of the pharmaceutical industry in Germany. Eur. Surg.
Res. 34, 203207 (2008).
14. Friedman, M. A. et al. The safety of newly approved
medicines: do recent market removals mean there is a
problem? JAMA 281, 17281734 (1999).
15. [No authors listed.] How a statin might destroy a drug
company. Lancet 361, 793 (2003).
16. European Medicines Agency. EPAR Avandia. European
Medicines Agency [online], http://www.ema.europa.eu/
docs/en_GB/document_library/EPAR_-_Assessment_
Report_-_Variation/human/000666/WC500021280.
pdf (2009).
17. Forslund, T. et al. Usage, risk and benefit of weight-
loss drugs in primary care. Journal of Obesity [online],
http://downloads.mts.hindawi.com/MTS-Files/JOBES/
papers/regular/459263.v2.pdf?AWSAccessKeyId=0C
X53QQSTHRYZZQRKA02&Expires=1305781317&
Signature=hAU9a2Tn5%2BKIOsuXtCdGERM0iqA%
3D (2010).
18. Center for Medical Technology Policy. Effectiveness
Guidance Document: Pragmatic Phase 3
Pharmaceutical Trials. Release Date: September 14,
2010. Center for Medical Technology Policy [online],
http://www.cmtpnet.org/cmtp-research/guidance-
documents/pragmatic-clinical-trials/PCT3EGD.pdf
(2010).
19. Petak, I. et al. Integrating molecular diagnostics into
anticancer drug discovery. Nature Rev. Drug Discov. 9,
523535 (2010).
20. Heerdink, E. R., Urquhart, J. & Leufkens, H. G.
Changes in prescribed doses after market introduction.
Pharmacoepidemiol. Drug Saf. 11, 447453 (2002).
21. Cross, J. et al. Postmarketing drug dosage changes of
499 FDA-approved new molecular entities, 1980
1999. Pharmacoepidemiol. Drug Saf. 11, 439446
(2002).
22. Trusheim, M. R. et al. Stratified medicine: strategic
and economic implications of combining drugs and
clinical biomarkers. Nature Rev. Drug Discov. 6,
287293 (2007).
23. European Medicines Agency. Herceptin EPAR.
europa.eu/docs/en_GB/document_library/EPAR_-_
Product_Information/human/000278/
WC500074922.pdf (Last updated 19 May 2011).
24. Barron, J. J., Cziraky, M. J., Weisman, T. & Hicks, D. G.
HER2 testing and subsequent trastuzumab treatment
for breast cancer in a managed care environment.
Oncologist 14, 760768 (2009).
25. European Medicines Agency. EPAR Ziagen. European
docs/en_GB/document_library/EPAR_-_Product_
Information/human/000252/WC500050343.pdf (Last
updated 18 Apr 2011).
26. Stocco, G., Crews, K. R. & Evans, W. E. Genetic
polymorphism of inosine-triphosphate-
pyrophosphatase influences mercaptopurine
metabolism and toxicity during treatment of acute
lymphoblastic leukemia individualized for
thiopurine-S-methyl-transferase status. Expert Opin.
Drug Saf. 9, 2337 (2010).
27. Zaza, G., Granata, S., Sallustio, F., Grandaliano, G. &
Schena, F. P. Pharmacogenomics: a new paradigm to
personalize treatments in nephrology patients. Clin.
Exp. Immunol.159, 268280 (2010).
28. ICH Expert Working Group. International Conference
on Harmonisation Guideline E5(R1): Ethnic Factors in
the Acceptability of Foreign Clinical Data. ICH
Harmonisation For Better Health [online], http://www.
ich.org/fileadmin/Public_Web_Site/ICH_Products/
Guidelines/Efficacy/E5_R1/Step4/E5_R1__Guideline.
pdf (1998).
29. Falagas, M. E. & Karageorgopoulos, D. E. Adjustment
of dosing of antimicrobial agents for bodyweight in
adults. Lancet 375, 248251 (2010).
30. Kirsch, I. et al. Initial severity and antidepressant
benefits: a meta-analysis of data submitted to the
Food and Drug Administration. PLoS Med. 5,
260268 (2008).
31. van Staa, T.-P., Leufkens, H. G., Zhang, B. & Smeeth, L.
A comparison of cost effectiveness using data from
randomized trials or actual clinical practice: selective
Cox-2 inhibitors as an example. PLoS Med. 6,
e1000194 (2009).
32. Greenblatt, D. J. Analysis of drug interactions
involving fruit beverages and organic anion-
transporting polypeptides. J.Clin. Pharmacol. 49,
14031407 (2009).
33. Bailey, D. G. Fruit juice inhibition of uptake transport:
a new type of fooddrug interaction. Br. J.Clin.
Pharmacol. 70, 645655 (2010).
34. Klaasen, R., Wijbrandts, C. A., Gerlag, D. M. & Tak,
P. P. Body mass index and clinical response to
infliximab in rheumatoid arthritis. Arthritis Rheum.
63, 359364 (2011).
35. US Government Accountability Office: Report to the
Ranking Member, Committee on Finance, US Senate.
Prescription drugs: FDAs oversight of the promotion
of drugs for off-label uses; July 2008. US Government
Accountability Office [online], http://www.gao.gov/new.
items/d08835.pdf (2008).
36. Radley. D.C., Finkelstein, S.N. & Stafford, R. S. Off-
label prescribing among office-based physicians. Arch.
Intern. Med. 166,10211026 (2006).
37. Jonville-Bra, A. P., Bra, F. & Autret-Lecaq, E. Are
incorrectly used drugs more frequently involved in
adverse drug reactions? A retrospective study. Eur.
J.Clin. Pharmacol. 61, 231236 (2005).
38. Cereza, G., Pedros, C., Garcia, N. & Laporte, J. R.
Topiramate in non-approved indications and acute
myopia or angle-closure glaucoma. Br. J.Clin.
Pharmacol. 60, 578579 (2005).
39. Kohn, L. T., Corrigan, J. M., & Donaldson, M. S. (eds)
To Err is Human: Building a Safer Health System
(National Academy Press, Washington DC, 2000).
40. Bonaccorso, S. & Sturchio, J. L. Perspectives from the
pharmaceutical industry. BMJ 327, 863864
(2003).
41. WHO. Adherence to long-term therapies: evidence for
action. World Health Organization [online], http://
apps.who.int/medicinedocs/en/d/Js4883e/ (2003).
42. Urquhart, J. The odds of the three nons when an aptly
prescribed medicine isnt working: non-compliance,
non-absorption, non-response. Br. J.Clin. Pharmacol.
54, 212220 (2002).
43. Horwitz, R. I. et al. Treatment adherence and risk of
death after a myocardial infarction. Lancet 336,
542545 (1990).
44. Cramer, J. A, Benedict, , Muszbek, N.,
Keskinaslan,A. & Khan, Z. M. The significance of
compliance and persistence in the treatment of
diabetes, hypertension and dyslipidaemia: a review.
Int. J.Clin. Pract. 62, 7687 (2008).
45. Vrijens, B., Gross, R. & Urquhart, J. The odds that
clinically unrecognised poor or partial adherence
confuses population pharmacokinetic/
PERSPECTI VES
nrd_3501_jul11.indd 505 20/06/2011 16:34
Srucurod
druq
inlormuion
GNCDGN
o-boulbcuro
rocord
o-boulbcuro
duubusos
Pos-murloinq
rosourcb und moniorinq
Comuorizod
bysiciun ordor
onry
Docision
suor sysom
Aroriuo
roscribinq
for example by the use of electronic event
monitors (see below). This added informa-
tion could be used to estimate, in secondary
analyses and by way of modelling and simu-
lation, the effect of incomplete adherence
on effect size and may yield information
on the forgiveness of drugs that is, the
sensitivity of drug effects to sporadic non-
adherence. Osterberg etal.
99
have recently
discussed past experiences with different
study designs and monitoring strategies,
including a design called placebo-substitu-
tion-for-active, which has been applied to
oral contraceptive agents. (Oral contracep-
tives are useful to illustrate different levels
of forgiveness, even within one class: the
original high-dose oral contraceptives were
more forgiving of incomplete patient adher-
ence but had a higher likelihood of causing
thrombotic events compared with the low-
dose oral contraceptive products introduced
later
99
.)
There is now a broad range of proven,
emerging or proposed interventions and
technologies that may successfully improve
adherence. These are generally motivational
and include financial incentives to patients,
payment systems that reward care providers
for better patient outcomes
97,98
and, increas-
ingly, interventions based on information
and communication technologies (ICT).
Directly observed therapy (DOT) pro-
vides an early example of the power of
achieving continuity of patients exposure
to drugs to close the efficacyeffectiveness
gap. Started in the 1990s in New York City,
DOT programmes, in which health work-
ers watch patients take their medications,
have made noticeable reductions in the
numbers of new cases of tuberculosis and in
the incidence of drug-resistant tuberculo-
sis
100
.However, DOT is resource-intensive
and is not practically feasible in most treat-
ment scenarios.
A simple ICT-based example is offered
by a recent study into the effects of a mobile
phone message on antiretroviral treatment
adherence in Kenya: HIV-infected adults
were randomized to a mobile phone short
message service (SMS) intervention or
standard care. Patients in the intervention
group received weekly SMS messages from
a clinic nurse and were required to respond
within 48hours. After 6 and 12months,
patients who received SMS support had
significantly improved drug adherence and
rates of viral suppression compared with the
control individuals
101
. It may be surprising
that such an infrequent and basic interven-
tion would be effective, and it was specu-
lated that the SMS intervention worked by
improving communication and rapport
between health providers and patients;
patients reported during the pilot phase
that it feels like someone cares (REF.102).
Whatever the reasons for its effect, this study
and others evaluating relatively low-cost self-
titration schemes
103
demonstrate the poten-
tial of inexpensive behavioural interventions
on treatmenteffect.
More consistent drug use may also be
supported by technologies such as drug
packaging devices for example, intelligent
pill caps that use light and sound which
can be followed by timely phone calls or text
messages to remind patients to adhere to a
prescription regimen, to order refills from
the pharmacy, and to provide feedback to
care providers
104
.
Complex ingestible electronic systems
are currently in development to enable elec-
tronically observed therapy. For instance,
an edible sensor (consisting of an integrated
circuit embedded in a solid drug dosage
form) is activated after ingestion by stom-
ach fluid and communicates a signal to a
wearable monitor attached to the patients
torso like an adhesive bandage. The monitor
transmits the drug signature, time of intake
and physiological data to a mobile phone,
enabling a feedback loop that includes
reminders to take a particular drug dosage
to patients themselves, to their caregivers or
to health-care providers
105
. To our knowl-
edge, none of these systems has been fully
assessed and important questions about the
long-term clinical safety and utility of ingest-
ible systems remain to be addressed, but the
approach might prove useful in thefuture.
Some adherence-enhancing interventions
are already in routine use, but their potential
is not fully exploited. Non-persistent and
non-adherent patients represent a loss of
income for companies and a loss to payers,
as their investment will not result in the
expected health gain, and low adherence to
rational drug therapy may give rise to higher
overall health-care costs
106
. We are there-
fore surprised by a lack of enthusiasm from
industry and payers of health care to explore
and adopt these new technologies more
aggressively. Regulators should encourage
and support drug developers to consider
adherence-promoting technologies during
pre-marketing development and as a part of
RMPs orREMSs.
Conclusions
The recent availability of electronic health-
care databases for observational studies of
treatment outcomes and the expanded use of
PCTs
107
is likely to reveal a growing number
of incidences in which effectiveness does not
match efficacy. Some effectiveness data may
challenge the actions of all concerned
industry, regulators, payers and health-care
providers.
Pharmacogenomics and phenotypic
biomarkers and, perhaps, new licensing
approaches are expected to reduce but not
eliminate variability of drug response. We
Figure 4 | The role of electronic drug information in reducing the efficacyeffectiveness gap. Most
building blocks of e-healthcare are becoming a reality, but the electronic drug information (e-label)
an electronic, structured format of the authorized drug information remains the missing link
63
.
e-label information will be useful at the individual patientprescriber level and at the health-care
systems level. At point of care, the e-label is linked with patient-specific information from the
e-healthcare record (which is critical in e-prescribing) and with computer physician order entry systems
and their integrated decision-support-systems. e-label information supports prescribing decisions by
alerts with warnings or contraindications that are relevant for a specific patient. At the level of the
health-care system, payers, regulators or health-care managers may link e-label information with
population e-healthcare databases for comparative effectiveness research, post-licensing effectiveness
and safety monitoring, and monitoring of quality of health care.
PERSPECTI VES
nrd_3501_jul11.indd 504 20/06/2011 16:34
have discussed the nonlinear interaction
of different sources of variability. It follows
that even large reductions in variance from
a single source may have counter-intuitively
small effects in reducing the overall variance
in drug response
46
. A more holistic effort
will therefore be required from all players
to tackle the efficacyeffectivenessgap.
Considering the time and cost of bringing
new drugs to market, it seems that attempts to
improve drug effectiveness in real life repre-
sent good value for money; this is a space with
low-hanging fruits that might substantially
improve public health. Revisiting Sir William
Oslers observation, when this potential is
realized, medicine might [indeed become]
a science and not an art.
Hans-Georg Eichler, Eric Abadie, Bruno Flamion,
Hubert Leufkens, Christian K.Schneider and Brigitte
Bloechl-Daum are at the European Medicines Agency,
London, UK.
Hans-Georg Eichler is also at the Massachusetts Institute
of Technology, Cambridge, Massachusetts, USA.
Eric Abadie is also at the European Medicines Agency
Committee for Medicinal Products for Human Use,
London, UK; and theGeneral Directorate Agence
Francaise de Securite Sanitaire des Produits de Sant
(AFSSAPS), Paris, France.
Alasdair Breckenridge is at the Medicines and
Healthcare Products Regulatory Agency, London, UK.
Bruno Flamion is also at the Agence Fdrale des
Mdicaments et des Produits de Sant, Brussels,
Belgium.
Lars L.Gustafsson is at the Division of Clinical
Pharmacology, Department of Laboratory Medicine,
Karolinska Institutet, Stockholm, Sweden.
Hubert Leufkens is also at the Utrecht Institute for
Pharmaceutical Sciences (UIPS), Utrecht,
The Netherlands.
Malcolm Rowland is at the Centre for Applied
Pharmacokinetic Research, School of Pharmacy and
Pharmaceutical Sciences, University of Manchester,
Manchester, UK.
Christian K.Schneider is also at the Paul-Ehrlich-Institut,
Langen, Germany; and the Twincore Centre for
Experimental and Clinical Infection Research,
Hannover, Germany.
Brigitte Bloechl-Daum is also at the Department of
Clinical Pharmacology, Medical University Vienna,
Vienna, Austria.
Correspondence to B.B.-D.
e-mail: brigitte.bloechl-daum@meduniwien.ac.at
doi: 10.1038/nrd3501
Disclaimer: The views expressed in this article are the
personal views of the authors and may not be
understood or quoted as being made on behalf of or
reflecting the position of the regulatory agencies,
health technology assessment bodies or other
organizations that the authors work for.
1. Danish Medicines Agency. Conclusions and
recommendations from the Pharmaceutical Forum.
Danish Medicines Agency [online], http://www.dkma.
dk/1024/visUKLSArtikel.asp?artikelID=17481
(2010).
2. Luce, B. R. et al. EBM, HTA, and CER: clearing the
confusion. Millbank Q. 88, 256276 (2010).
3. European Medicines Agency. European Medicines
Agency recommends measures to manage
contamination of heparin-containing medicines.
europa.eu/ema/index.jsp?curl=pages/news_and_
events/news/2009/11/news_detail_000315.
jsp&murl=menus/news_and_events/news_and_
events.jsp&mid=WC0b01ac058004d5c1&jsenabled
=true (2008).
4. European Medicines Agency. Studies assessed by the
EMEA indicate no increased risk of developing cancer
for patients who have taken Viracept contaminated
with ethyl mesilate. European Medicines Agency
[online]. http://www.ema.europa.eu/ema/index.
jsp?curl=pages/news_and_events/news/2009/11/
news_detail_000292.jsp&murl=menus/news_and_
events/news_and_events.
jsp&mid=WC0b01ac058004d5c1 (2008).
5. Dal Pan, G. J., Blackburn, S. & Karwoski, C. in
Textbook of Pharmacoepidemiology (eds Strom, B. L.
& Kimmel, S. E.) (John Wiley & Sons, New Jersey), (in
the press).
6. Pocock, S. J. & Lubsen, J. More on subgroup analyses
in clinical trials. N.Engl. J.Med. 358, 2076 (2008).
7. Ingelman-Sundberg, M. Pharmacogenetics: an
opportunity for a safer and more efficient
pharmacotherapy. J. Intern. Med. 250, 186200
(2001).
Drug Discov. 7, 807817 (2008).
9. Pharoa, P. D. & Hollingworth, W. Cost effectiveness of
lowering cholesterol concentration with statins in
patients with and without pre-existing coronary heart
disease: life table method applied to health authority
population. BMJ 312, 14431448 (1996).
10. Poole, S. G. & Dooley, M. J. Off-label prescribing in
oncology. Support Care Cancer 12, 302305 (2004).
11. Hsien, L. et al. Off-label drug use among hospitalised
children: identifying areas with the highest need for
research. Pharm. World Sci. 30, 497502 (2008).
12. [No authors listed.] Guidance for off-label use of drugs.
Lancet Neurol. 7, 285 (2008).
13. Schosser, R. Risk/benefit evaluation of drugs: the role
of the pharmaceutical industry in Germany. Eur. Surg.
Res. 34, 203207 (2008).
14. Friedman, M. A. et al. The safety of newly approved
medicines: do recent market removals mean there is a
problem? JAMA 281, 17281734 (1999).
15. [No authors listed.] How a statin might destroy a drug
company. Lancet 361, 793 (2003).
16. European Medicines Agency. EPAR Avandia. European
docs/en_GB/document_library/EPAR_-_Assessment_
Report_-_Variation/human/000666/WC500021280.
pdf (2009).
17. Forslund, T. et al. Usage, risk and benefit of weight-
loss drugs in primary care. Journal of Obesity [online],
http://downloads.mts.hindawi.com/MTS-Files/JOBES/
papers/regular/459263.v2.pdf?AWSAccessKeyId=0C
X53QQSTHRYZZQRKA02&Expires=1305781317&
Signature=hAU9a2Tn5%2BKIOsuXtCdGERM0iqA%
3D (2010).
18. Center for Medical Technology Policy. Effectiveness
Guidance Document: Pragmatic Phase 3
Pharmaceutical Trials. Release Date: September 14,
2010. Center for Medical Technology Policy [online],
http://www.cmtpnet.org/cmtp-research/guidance-
documents/pragmatic-clinical-trials/PCT3EGD.pdf
(2010).
19. Petak, I. et al. Integrating molecular diagnostics into
anticancer drug discovery. Nature Rev. Drug Discov. 9,
523535 (2010).
20. Heerdink, E. R., Urquhart, J. & Leufkens, H. G.
Changes in prescribed doses after market introduction.
Pharmacoepidemiol. Drug Saf. 11, 447453 (2002).
21. Cross, J. et al. Postmarketing drug dosage changes of
499 FDA-approved new molecular entities, 1980
1999. Pharmacoepidemiol. Drug Saf. 11, 439446
(2002).
22. Trusheim, M. R. et al. Stratified medicine: strategic
and economic implications of combining drugs and
clinical biomarkers. Nature Rev. Drug Discov. 6,
287293 (2007).
23. European Medicines Agency. Herceptin EPAR.
europa.eu/docs/en_GB/document_library/EPAR_-_
Product_Information/human/000278/
WC500074922.pdf (Last updated 19 May 2011).
24. Barron, J. J., Cziraky, M. J., Weisman, T. & Hicks, D. G.
HER2 testing and subsequent trastuzumab treatment
for breast cancer in a managed care environment.
Oncologist 14, 760768 (2009).
25. European Medicines Agency. EPAR Ziagen. European
docs/en_GB/document_library/EPAR_-_Product_
Information/human/000252/WC500050343.pdf (Last
updated 18 Apr 2011).
26. Stocco, G., Crews, K. R. & Evans, W. E. Genetic
polymorphism of inosine-triphosphate-
pyrophosphatase influences mercaptopurine
metabolism and toxicity during treatment of acute
lymphoblastic leukemia individualized for
thiopurine-S-methyl-transferase status. Expert Opin.
Drug Saf. 9, 2337 (2010).
27. Zaza, G., Granata, S., Sallustio, F., Grandaliano, G. &
Schena, F. P. Pharmacogenomics: a new paradigm to
personalize treatments in nephrology patients. Clin.
Exp. Immunol.159, 268280 (2010).
28. ICH Expert Working Group. International Conference
on Harmonisation Guideline E5(R1): Ethnic Factors in
the Acceptability of Foreign Clinical Data. ICH
Harmonisation For Better Health [online], http://www.
ich.org/fileadmin/Public_Web_Site/ICH_Products/
Guidelines/Efficacy/E5_R1/Step4/E5_R1__Guideline.
pdf (1998).
29. Falagas, M. E. & Karageorgopoulos, D. E. Adjustment
of dosing of antimicrobial agents for bodyweight in
adults. Lancet 375, 248251 (2010).
30. Kirsch, I. et al. Initial severity and antidepressant
benefits: a meta-analysis of data submitted to the
Food and Drug Administration. PLoS Med. 5,
260268 (2008).
31. van Staa, T.-P., Leufkens, H. G., Zhang, B. & Smeeth, L.
A comparison of cost effectiveness using data from
randomized trials or actual clinical practice: selective
Cox-2 inhibitors as an example. PLoS Med. 6,
e1000194 (2009).
32. Greenblatt, D. J. Analysis of drug interactions
involving fruit beverages and organic anion-
transporting polypeptides. J.Clin. Pharmacol. 49,
14031407 (2009).
33. Bailey, D. G. Fruit juice inhibition of uptake transport:
a new type of fooddrug interaction. Br. J.Clin.
Pharmacol. 70, 645655 (2010).
34. Klaasen, R., Wijbrandts, C. A., Gerlag, D. M. & Tak,
P. P. Body mass index and clinical response to
infliximab in rheumatoid arthritis. Arthritis Rheum.
63, 359364 (2011).
35. US Government Accountability Office: Report to the
Ranking Member, Committee on Finance, US Senate.
Prescription drugs: FDAs oversight of the promotion
of drugs for off-label uses; July 2008. US Government
Accountability Office [online], http://www.gao.gov/new.
items/d08835.pdf (2008).
36. Radley. D.C., Finkelstein, S.N. & Stafford, R. S. Off-
label prescribing among office-based physicians. Arch.
Intern. Med. 166,10211026 (2006).
37. Jonville-Bra, A. P., Bra, F. & Autret-Lecaq, E. Are
incorrectly used drugs more frequently involved in
adverse drug reactions? A retrospective study. Eur.
J.Clin. Pharmacol. 61, 231236 (2005).
38. Cereza, G., Pedros, C., Garcia, N. & Laporte, J. R.
Topiramate in non-approved indications and acute
myopia or angle-closure glaucoma. Br. J.Clin.
Pharmacol. 60, 578579 (2005).
39. Kohn, L. T., Corrigan, J. M., & Donaldson, M. S. (eds)
To Err is Human: Building a Safer Health System
(National Academy Press, Washington DC, 2000).
40. Bonaccorso, S. & Sturchio, J. L. Perspectives from the
pharmaceutical industry. BMJ 327, 863864
(2003).
41. WHO. Adherence to long-term therapies: evidence for
action. World Health Organization [online], http://
apps.who.int/medicinedocs/en/d/Js4883e/ (2003).
42. Urquhart, J. The odds of the three nons when an aptly
prescribed medicine isnt working: non-compliance,
non-absorption, non-response. Br. J.Clin. Pharmacol.
54, 212220 (2002).
43. Horwitz, R. I. et al. Treatment adherence and risk of
death after a myocardial infarction. Lancet 336,
542545 (1990).
44. Cramer, J. A, Benedict, , Muszbek, N.,
Keskinaslan,A. & Khan, Z. M. The significance of
compliance and persistence in the treatment of
diabetes, hypertension and dyslipidaemia: a review.
Int. J.Clin. Pract. 62, 7687 (2008).
45. Vrijens, B., Gross, R. & Urquhart, J. The odds that
clinically unrecognised poor or partial adherence
confuses population pharmacokinetic/
PERSPECTI VES
nrd_3501_jul11.indd 505 20/06/2011 16:34
Srucurod
druq
inlormuion
GNCDGN
o-boulbcuro
rocord
o-boulbcuro
duubusos
Pos-murloinq
rosourcb und moniorinq
Comuorizod
bysiciun ordor
onry
Docision
suor sysom
Aroriuo
roscribinq
for example by the use of electronic event
monitors (see below). This added informa-
tion could be used to estimate, in secondary
analyses and by way of modelling and simu-
lation, the effect of incomplete adherence
on effect size and may yield information
on the forgiveness of drugs that is, the
sensitivity of drug effects to sporadic non-
adherence. Osterberg etal.
99
have recently
discussed past experiences with different
study designs and monitoring strategies,
including a design called placebo-substitu-
tion-for-active, which has been applied to
oral contraceptive agents. (Oral contracep-
tives are useful to illustrate different levels
of forgiveness, even within one class: the
original high-dose oral contraceptives were
more forgiving of incomplete patient adher-
ence but had a higher likelihood of causing
thrombotic events compared with the low-
dose oral contraceptive products introduced
later
99
.)
There is now a broad range of proven,
emerging or proposed interventions and
technologies that may successfully improve
adherence. These are generally motivational
and include financial incentives to patients,
payment systems that reward care providers
for better patient outcomes
97,98
and, increas-
ingly, interventions based on information
and communication technologies (ICT).
Directly observed therapy (DOT) pro-
vides an early example of the power of
achieving continuity of patients exposure
to drugs to close the efficacyeffectiveness
gap. Started in the 1990s in New York City,
DOT programmes, in which health work-
ers watch patients take their medications,
have made noticeable reductions in the
numbers of new cases of tuberculosis and in
the incidence of drug-resistant tuberculo-
sis
100
.However, DOT is resource-intensive
and is not practically feasible in most treat-
ment scenarios.
A simple ICT-based example is offered
by a recent study into the effects of a mobile
phone message on antiretroviral treatment
adherence in Kenya: HIV-infected adults
were randomized to a mobile phone short
message service (SMS) intervention or
standard care. Patients in the intervention
group received weekly SMS messages from
a clinic nurse and were required to respond
within 48hours. After 6 and 12months,
patients who received SMS support had
significantly improved drug adherence and
rates of viral suppression compared with the
control individuals
101
. It may be surprising
that such an infrequent and basic interven-
tion would be effective, and it was specu-
lated that the SMS intervention worked by
improving communication and rapport
between health providers and patients;
patients reported during the pilot phase
that it feels like someone cares (REF.102).
Whatever the reasons for its effect, this study
and others evaluating relatively low-cost self-
titration schemes
103
demonstrate the poten-
tial of inexpensive behavioural interventions
on treatmenteffect.
More consistent drug use may also be
supported by technologies such as drug
packaging devices for example, intelligent
pill caps that use light and sound which
can be followed by timely phone calls or text
messages to remind patients to adhere to a
prescription regimen, to order refills from
the pharmacy, and to provide feedback to
care providers
104
.
Complex ingestible electronic systems
are currently in development to enable elec-
tronically observed therapy. For instance,
an edible sensor (consisting of an integrated
circuit embedded in a solid drug dosage
form) is activated after ingestion by stom-
ach fluid and communicates a signal to a
wearable monitor attached to the patients
torso like an adhesive bandage. The monitor
transmits the drug signature, time of intake
and physiological data to a mobile phone,
enabling a feedback loop that includes
reminders to take a particular drug dosage
to patients themselves, to their caregivers or
to health-care providers
105
. To our knowl-
edge, none of these systems has been fully
assessed and important questions about the
long-term clinical safety and utility of ingest-
ible systems remain to be addressed, but the
approach might prove useful in thefuture.
Some adherence-enhancing interventions
are already in routine use, but their potential
is not fully exploited. Non-persistent and
non-adherent patients represent a loss of
income for companies and a loss to payers,
as their investment will not result in the
expected health gain, and low adherence to
rational drug therapy may give rise to higher
overall health-care costs
106
. We are there-
fore surprised by a lack of enthusiasm from
industry and payers of health care to explore
and adopt these new technologies more
aggressively. Regulators should encourage
and support drug developers to consider
adherence-promoting technologies during
pre-marketing development and as a part of
RMPs orREMSs.
Conclusions
The recent availability of electronic health-
care databases for observational studies of
treatment outcomes and the expanded use of
PCTs
107
is likely to reveal a growing number
of incidences in which effectiveness does not
match efficacy. Some effectiveness data may
challenge the actions of all concerned
industry, regulators, payers and health-care
providers.
Pharmacogenomics and phenotypic
biomarkers and, perhaps, new licensing
approaches are expected to reduce but not
eliminate variability of drug response. We
Figure 4 | The role of electronic drug information in reducing the efficacyeffectiveness gap. Most
building blocks of e-healthcare are becoming a reality, but the electronic drug information (e-label)
an electronic, structured format of the authorized drug information remains the missing link
63
.
e-label information will be useful at the individual patientprescriber level and at the health-care
systems level. At point of care, the e-label is linked with patient-specific information from the
e-healthcare record (which is critical in e-prescribing) and with computer physician order entry systems
and their integrated decision-support-systems. e-label information supports prescribing decisions by
alerts with warnings or contraindications that are relevant for a specific patient. At the level of the
health-care system, payers, regulators or health-care managers may link e-label information with
population e-healthcare databases for comparative effectiveness research, post-licensing effectiveness
and safety monitoring, and monitoring of quality of health care.
PERSPECTI VES
nrd_3501_jul11.indd 504 20/06/2011 16:34
N
u
m
b
o
r

o
l

d
r
u
q
s
35
30
25
20
15
10
5
0
2004 2006 2007 2008 2009 2010 2003 2005
Your
lND NDA
14
1
20
4
30
11
21
2 2 2
1
0
24
21
21
24
applications (NDAs) and investigational new
drug applications (INDs) were approved in
China between 2003 and 2010. The novel
pharmaceuticals discussed in this article only
include chemical drugs in classes 1.1 and
1.2 and biological drugs in class 1, which are
defined by the SFDA as not being previously
approved for marketing as a drug anywhere
else in the world. Thus, the scope of novel
drugs in this article is narrower than that of
the new chemical entities (NCEs) used by the
US Food and Drug Administration (FDA).
For example, if a drug was first approved by
the FDA or the European Medicines Agency,
it would not be qualified as a class 1 new drug
by the SFDA when its approval is later sought
in China. Multinational pharmaceutical
companies usually market their drugs in
developed countries first, and in our survey,
all the innovative drugs approved as class
1 by the SFDA are from domestic Chinese
pharmaceutical companies. Consequently,
the class 1 approvals by the SFDA reflect
the status of innovative drug development
solely in domestic Chinese pharmaceutical
companies. Additionally, traditional Chinese
medicine and vaccines are not discussed here.
Analysis
In the period analysed, ~25 drug candidates
were approved for entry into clinical trials
(that is, an IND was granted) and an average
of four drugs were approved for marketing
per year (FIG. 1). However, there was a
significant reduction in the number of drugs
approved for marketing since 2006, with less
than two approvals per year in comparison
with the preceding years that ended with
11 approvals in 2005. This is primarily due
to the introduction by the SFDA in 2007 of
much more stringent rules and regulations
regarding new drug approval and
registration.
A total of 187 novel therapeutics are
currently in clinical trials. Nearly two-thirds
of the therapeutics are in Phase I trials, with
those in Phase II and III trials accounting for
19% and 22%, respectively (FIG. 2a). As shown
in FIG. 2b, oncology is the most common
therapeutic area (32% of therapeutics
analysed), followed by infectious diseases
(17%) and cardiovascular diseases (10%).
With a population of 1.3 billion people and
a rapidly expanding economy, China has
recently risen to become the third largest
pharmaceutical market globally, and it has
been predicted that this market will grow
by 2527% to a value of more than US$50
billion in 2011 (REF. 1). Although many of
the drugs in the current market are either
generic versions or developed outside
China, several multinational pharmaceutical
companies have now located research and
development (R&D) centres in China, and
Chinese pharmaceutical companies are
increasingly focusing on innovative drug
R&D. Furthermore, the Chinese government
implemented a special drug R&D funding
programme in which $2.7 billion was invested
from 2008 to 2010, with another $6 billion to
follow in the next 5 years.
However, information on the output of
innovative drug R&D in China is limited.
With the aim of addressing this issue, we
have collected and analysed information
from the Chinese State Food and Drug
Administration (SFDA) and the Center for
Drug Evaluation of the SFDA (CDE) for all
novel pharmaceuticals for which new drug
FROM THE ANALYSTS COUCH
Innovative drug R&D in China
Jingzong Qi, Qingli Wang, Zhenhang Yu, Xin Chen and Fengshan Wang
Arne Jacobsen: 1969.
Dansk Mobelkunst
Gallery www.dmk.dk
Figure 1 | Annual number of approved INDs
and NDAs for innovative drugs from Chinese
companies. Data were collected from the
website of the Chinese State Food and Drug
Administration (SFDA) and the Center for Drug
Evaluation of the SFDA. IND, investigational new
drug application; NDA, new drug application.
Bearing in mind the importance of
patent protection in drug R&D as well as the
evolution of Chinese patent law in recent
years, we also analysed the patenting of the
investigational therapeutics in China. Patents
for pharmaceuticals are divided into two
categories: compound patents and secondary
patents, which include preparation, detection,
pharmaceutical composition and usage.
Out of 187 investigational drugs, 70 have
compound patent protection in China,
whereas 23 have compound patent protection
in the United States and 16 in Europe (FIG. 2c).
We also investigated the characteristics of
those novel therapeutics that had patent
protection in either the United States or
Europe, which are presented in TABLE 1.
Outlook
The increased investment by the
Chinese government and multinational
pharmaceutical companies, as well as
other improvements (discussed below),
are creating a stronger environment for
innovative drug R&D in China. First, in the
past two decades, the Chinese regulatory
system has undergone a systematic
transformation to adapt to the emergence of
more INDs. The first drug administration
law in China was enacted in 1985, and
there have been four major amendments
since, with the latest one enacted in 2007.
As mentioned above, the SFDA (which
itself was founded in 2003) has introduced
more robust regulations regarding new drug
approval and registration, and to improve
transparency and efficiency, which could
pave the way for the emergence of more
innovative drugs in the long term. For
example, the registration status of an IND
or NDA is publicly available, and applicants
have easy access to all information regarding
the approval process for their drugs. Local
agencies of the SFDA are authorized to
conduct preliminary approval procedures
to increase efficiency. Companies that
provide false information or samples will be
penalized and barred from submitting NDAs
for up to 3 years.
Patent protection is a second factor that
is essential in innovative drug development.
The current Chinese patent law was enacted
doi:10.1038/nrd3435
NEWS & ANALYSI S
NATURE REVIEWS | DRUG DISCOVERY VOLUME 10 | MAY 2011 | 333
nrd_3435_may11.indd 333 14/04/2011 11:49
pharmacodynamic analyses. Basic Clin. Pharmacol.
Toxicol. 96, 225227 (2005).
46. Urquhart, J. Getting a handle on why good drugs
sometimes dont work. J.R.Coll. Physicians Edinb.
34, 9598 (2004).
47. Centers for Disease Control and Prevention.
Achievements in Public Health, 19001999 Family
Planning. MMWR Morb. Mortal. Wkly. Rep. 48,
10731080 (1999).
48. Vrijens, B. et al. Modellling the association between
adherence and viral load in HIV-infected patients. Stat.
Med. 24, 27192731 (2005).
49. Vrijens, B. & Urquhart, J. Patient adherence to
prescribed antimicrobial drug dosing regimens.
J.Antimicrob. Chemother. 55, 616627 (2005).
50. Olfson, M., West, J. C., Wilk, J. E. & Marcus, S. Factors
affecting the effectiveness of clinical decisions in
treating schizophrenia. in Proc. of the American
Psychiatric Assoc. 156th Annual Meeting
(1722May 2003; San Francisco, California, USA;
Abstract S28C).
51. Vrijens, B., Vincze, G., Kristanto, P., Urquhart, J. &
Burnier, M. Adherence to prescribed antihypertensive
drug treatments: longitudinal study of electronically
compiled dosing histories. BMJ 336, 11141117
(2008).
52. Vincent, O. et al. Effect of concomitant CYP2D6
inhibitor use and tamoxifen adherence on breast
cancer recurrence in early-stage breast cancer. J. Clin.
Oncol. 28, 24232429 (2010).
53. Saevarsdottir, S. et al. Patients with early rheumatoid
arthritis who smoke are less likely to respond to
treatment with methotrexate and tumor necrosis
factor inhibitors. Arthtitis Rheum. 63. 2636 (2011).
54. Harter, J. G. & Peck, C. C. Chronobiology: suggestions
for integrating it into drug development. Ann.
N.Y.Acad. Sci. 618, 563571 (1991).
55. Rothwell, P. M. Factors that can affect the external
validity of randomised controlled trials. PLoS Clin.
Trials 1, e9 (2006).
56. Thorpe, K. E. et al. A pragmatic-explanatory
continuum indicator summary (PRECIS): a tool to help
trial designers. CMAJ 180, e47e57 (2009).
57. Temple, R. Enrichment of clinical study populations.
58. Goldstein, J. Why Medicare Pays so Much For Cancer
Drugs. European Medicines Agency. Wall Street
Journal [online], http://blogs.wsj.com/
health/2009/01/27/why-medicare-pays-so-much-for-ca
ncer-drugs/tab/comments/ (2009).
59. Barbui, C. & Garattini, S. Regulatory policies on
medicines for psychiatric disorders: is Europe on
target? Br. J.Psychiatry 190, 9193 (2007).
60. Woosley, R. L. & Rice, G. A new system for moving drugs
to market. Issues Sci. Technol. XXI, 6368 (2005).
61. Eichler, H. G., Pignatti, F., Flamion, B., Leufkens, H. &
Breckenridge, A. Balancing early market access to new
drugs with the need for benefit/risk data: a mounting
dilemma. Nature Rev. Drug Discov. 7, 818826
(2008).
62. [No authors listed.] Road map to 2015. European
docs/en_GB/document_library/Report/2011/01/
WC500101373.pdf (2010).
63. Maxwell, S., Eichler, H. G., Bucsics, A., Haefeli, W. E. &
Gustafsson, L.L. on behalf of the e-SPC consortium.
e-SPC delivering drug information in the 21st
century: developing new approaches to deliver drug
information to prescribers. Br. J.Clin. Pharmacol. 6Apr
2011 (doi:10.1111/j.136521252011.03981.x.).
64. Urquhart, J. Patient noncompliance with drug
regimens: measurement, clinical correlates, economic
impact. Eur. Heart J. 17 (Suppl. A), 815 (1996).
65. Leufkens, H. G. & Urquhart, J. Variability in patterns of
drug usage. J.Pharm. Pharmacol. 46 (Suppl. 1),
433437 (1994).
66. McQuay, H. J. & Moore, R. A. Using numerical results
from systematic reviews in clinical practice. Ann.
Intern. Med. 126, 712720 (1997).
67. Woodcock J. Assessing the clinical utility of diagnostics
used in drug therapy. Clin. Pharmacol. Ther. 88,
765773 (2010).
68. Mok, T. S. et al. Gefitinib or carboplatin-paclitaxel in
pulmonary adenocarcinoma. N.Engl. J. Med. 361,
947957 (2009).
69. Donnelly, L. A. et al. A paucimorphic variant in the
HMG-CoA reductase gene is associated with lipid-
lowering response to statin treatment in diabetes: a
GoDARTS study. Pharmacogenet. Genomics 18,
10211026 (2008).
70. Flockhart, D. A. Pharmacogenetic testing of CYP2C9
and VKORC1 alleles for warfarin. Genet. Med. 10,
139150 (2008).
71. Ginsburg, G. S. & Voora, D. The long and winding road
to warfarin pharmacogenetic testing. J.Am. Coll.
Cardiol. 55, 28132815 (2010).
72. Kamali, F. & Wynne, H. Pharmacogenetics of warfarin.
Annu. Rev. Med. 61, 6375 (2010).
73. Schilsky, R. L. Personalised medicine in oncology.
Drug Discov. 7, 807817 (2008).
75. European Medicines Agency. Reflection paper on
co-development of pharmacogenomic biomarkers and
assays in the context of drug development. European
docs/en_GB/document_library/Scientific_
guideline/2010/07/WC500094445.pdf (2010).
76. FDA. International Pharmaceutical Regulatory
Monitor: US FDA draft guidance explains new drug
development tools. FDA News [online], http://www.
fdanews.com/newsletter/article?issueId=14226&artic
leId=131972 (2010).
77. European Medicines Agency. Qualification of novel
methodologies for drug development: guidance to
applicants. European Medicines Agency [online],
http://www.ema.europa.eu/docs/en_GB/document_
library/Regulatory_and_procedural_
guideline/2009/10/WC500004201.pdf (2009).
78. McClellan, M. et al. An accelerated pathway for
targeted cancer therapies. Nature Rev. Drug Discov.
10, 7980 (2011).
79. [No authors listed.] Rethinking therapeutic cancer
vaccines. Nature Rev. Drug Discov. 8, 685686
(2009).
80. Hampel, H. et al. Biomarkers for Alzheimers disease
therapeutic trials. Prog. Neurobiol. 2Dec 2010
(doi:10.1016/j.pneurobio.2010.11.005).
81. Falissard, B. et al. Relative effectiveness assessment of
listed drugs (REAL): a new method for an early
comparison of the effectiveness of approved health
technologies. Int. J.Technol. Assess Health Care 26,
124130 (2010).
82. Birkett, D. et al. Clinical pharmacology in research,
teaching and health care: considerations by IUPHAR,
the Inter-national Union of Basic and Clinical
Pharmacology. Basic Clin. Pharmacol. Toxicol. 107,
531559 (2010).
83. Pocock, S. J., Assmann, S. E., Enos, L. E. & Kasten, L. E.
Subgroup analysis, covariate adjustment and
baseline comparisons in clinical trial reporting: current
practice and problems. Stat. Med. 21, 29172930
(2002).
84. Aarons, L. et al. Role of modelling and simulation in
Phase I drug development. Eur. J.Pharm. Sci. 13,
115122 (2001).
85. Manolis, E. & Pons, G. Proposals for model-based
paediatric medicinal development within the current
European Union regulatory framework. Br. J.Clin.
Pharmacol. 68, 493501 (2009).
86. [No authors listed.] How to reduce prescribing errors.
Lancet 374, 1945 (2009).
87. [No authors listed.] Evidence-based Practice Centres:
synthesizing scientific evidence to improve quality and
effectiveness in health care. Agency for Healthcare
Research and Quality [online], http://www.ahrq.gov/
clinic/epc/ (2011).
88. Bahri, P. Public pharmacovigilance communication:
a process calling for evidence-based, objective-driven
strategies. Drug Saf. 33,10651079 (2010).
89. Smalley, W. et al. Contraindicated use of cisapride:
impact of food and drug administration regulatory
action. JAMA 284, 30363039 (2000).
90. John, J. et al. HER2 testing and subsequent
trastuzumab treatment for breast cancer in a
managed care environment. Oncologist 14, 760768
(2009).
91. Gustafsson, L. L. et al. The Wise List a
comprehensive concept to select, communicate and
achieve recommendations of essential drugs in
ambulatory care in Stockholm. Basic Clin. Pharmacol.
Toxicol.108, 224233 (2011).
92. Shea, S. & Hripcsak, G. Accelerating the use of
electronic health records in physician practices.
N.Engl. J.Med. 362, 192195 (2010).
93. Sjborg, B. et al. Design and implementation of a
point-of-care computerized system for drug therapy in
Stockholm metropolitan health region-bridging the
gap between knowledge and practice. Int. J.Med.
Inform. 76, 497506 (2007).
94. Classen, D. C., Avery, A. J. & Bates, D. W. Evaluation
and certification of computerized provider order entry
systems. J.Am. Med. Inform. Assoc. 14, 4855
(2007).
95. Bttiger, Y. et al. SFINX a drugdrug interaction
database desgined for clinical decision support
systems. Eur. J.Clin. Pharmacol. 95, 627633
(2009).
96. Wettermark, B. et al. The new Swedish Prescribed
Drug Registeropportunities for
pharmacoepidemiological research and experience
from the first six months. Pharmacoepidemiol. Drug
Saf. 16, 726735 (2007).
97. Cutler, D. & Everett, W. Thinking outside the pillbox
medication adherence as a priority for health care
reform. N.Engl. J.Med. 362, 15531555
(2010).
98. Modi, A. C. & Quittner, A. L. Barriers to treatment
adherence for children with cystic fibrosis and asthma:
what gets in the way? J.Pediatr. Psychol. 31,
846858 (2006).
99. Osterberg, L. G., Urquhart, J. & Blaschke, T. F.
Understanding forgiveness: minding and mining the
gaps between pharmacokinetics and therapeutics.
100. Frieden, T. R., Fujiwara, P. I., Washko, R. M. &
Hamburg, M. A. Tuberculosis in New York City
turning the tide. N.Engl. J.Med. 333, 229233
(1995).
101. Lester, R. T. et al. Effects of a mobile phone short
message service on antiretroviral treatment adherence
in Kenya (WelTel Kenya1): a randomised trial. Lancet
376, 18381845 (2010).
102. Chi, B. H. & Stringer, J. S. A. Mobile phones to
improve HIV treatment adherence. Lancet 376,
18071808 (2010).
103. Ogedegbe G. Self-titration for treatment of
uncomplicated hypertension. Lancet 376, 144146
(2010).
104. Singer, E. Message from a pill bottle. Technology
Review [online], http://www.technologyreview.com/
business/26414/?p1=A4 (2010).
105. Au-Yeung, K. Y. et al. A networked system for self-
management of drug therapy and wellness. ACM
Digital Library [online], http://delivery.acm.
org/10.1145/1930000/1921083/p1-au-yeung.pdf?ke
y1=1921083&key2=9929465921&coll=DL&dl=
ACM&CFID=7014173&CFTOKEN=96965669
(2010).
106. Sokol, M. C., McGuigan, K. A., Verbrugge, R. R. &
Epstein, R. S. Impact of medication adherence on
hospitalization risk and healthcare cost. Med. Care
43, 521530 (2005).
107. Ware, J. H. & Hamel, M. B. Pragmatic trials guide
to better patient care? N.Engl. J. Med. 364,
16851687 (2011).
108. Hughes, B. 2009 FDA drug approvals. Nature Rev.
Drug Discov. 9, 8992 (2010).
109. Eichler, H. G., Aronsson, B., Abadie, E. & Salmonson, T.
New drug approval success rate in Europe
in 2009. Nature Rev. Drug Discov. 9, 355356
(2010).
PERSPECTI VES
nrd_3501_jul11.indd 506 20/06/2011 16:34
N
u
m
b
o
r

o
l

d
r
u
q
s
35
30
25
20
15
10
5
0
2004 2006 2007 2008 2009 2010 2003 2005
Your
lND NDA
14
1
20
4
30
11
21
2 2 2
1
0
24
21
21
24
applications (NDAs) and investigational new
drug applications (INDs) were approved in
China between 2003 and 2010. The novel
pharmaceuticals discussed in this article only
include chemical drugs in classes 1.1 and
1.2 and biological drugs in class 1, which are
defined by the SFDA as not being previously
approved for marketing as a drug anywhere
else in the world. Thus, the scope of novel
drugs in this article is narrower than that of
the new chemical entities (NCEs) used by the
US Food and Drug Administration (FDA).
For example, if a drug was first approved by
the FDA or the European Medicines Agency,
it would not be qualified as a class 1 new drug
by the SFDA when its approval is later sought
in China. Multinational pharmaceutical
companies usually market their drugs in
developed countries first, and in our survey,
all the innovative drugs approved as class
1 by the SFDA are from domestic Chinese
pharmaceutical companies. Consequently,
the class 1 approvals by the SFDA reflect
the status of innovative drug development
solely in domestic Chinese pharmaceutical
companies. Additionally, traditional Chinese
medicine and vaccines are not discussed here.
Analysis
In the period analysed, ~25 drug candidates
were approved for entry into clinical trials
(that is, an IND was granted) and an average
of four drugs were approved for marketing
per year (FIG. 1). However, there was a
significant reduction in the number of drugs
approved for marketing since 2006, with less
than two approvals per year in comparison
with the preceding years that ended with
11 approvals in 2005. This is primarily due
to the introduction by the SFDA in 2007 of
much more stringent rules and regulations
regarding new drug approval and
registration.
A total of 187 novel therapeutics are
currently in clinical trials. Nearly two-thirds
of the therapeutics are in Phase I trials, with
those in Phase II and III trials accounting for
19% and 22%, respectively (FIG. 2a). As shown
in FIG. 2b, oncology is the most common
therapeutic area (32% of therapeutics
analysed), followed by infectious diseases
(17%) and cardiovascular diseases (10%).
With a population of 1.3 billion people and
a rapidly expanding economy, China has
recently risen to become the third largest
pharmaceutical market globally, and it has
been predicted that this market will grow
by 2527% to a value of more than US$50
billion in 2011 (REF. 1). Although many of
the drugs in the current market are either
generic versions or developed outside
China, several multinational pharmaceutical
companies have now located research and
development (R&D) centres in China, and
Chinese pharmaceutical companies are
increasingly focusing on innovative drug
R&D. Furthermore, the Chinese government
implemented a special drug R&D funding
programme in which $2.7 billion was invested
from 2008 to 2010, with another $6 billion to
follow in the next 5 years.
However, information on the output of
innovative drug R&D in China is limited.
With the aim of addressing this issue, we
have collected and analysed information
from the Chinese State Food and Drug
Administration (SFDA) and the Center for
Drug Evaluation of the SFDA (CDE) for all
novel pharmaceuticals for which new drug
FROM THE ANALYSTS COUCH
Innovative drug R&D in China
Jingzong Qi, Qingli Wang, Zhenhang Yu, Xin Chen and Fengshan Wang
Arne Jacobsen: 1969.
Dansk Mobelkunst
Gallery www.dmk.dk
Figure 1 | Annual number of approved INDs
and NDAs for innovative drugs from Chinese
companies. Data were collected from the
website of the Chinese State Food and Drug
Administration (SFDA) and the Center for Drug
Evaluation of the SFDA. IND, investigational new
drug application; NDA, new drug application.
Bearing in mind the importance of
patent protection in drug R&D as well as the
evolution of Chinese patent law in recent
years, we also analysed the patenting of the
investigational therapeutics in China. Patents
for pharmaceuticals are divided into two
categories: compound patents and secondary
patents, which include preparation, detection,
pharmaceutical composition and usage.
Out of 187 investigational drugs, 70 have
compound patent protection in China,
whereas 23 have compound patent protection
in the United States and 16 in Europe (FIG. 2c).
We also investigated the characteristics of
those novel therapeutics that had patent
protection in either the United States or
Europe, which are presented in TABLE 1.
Outlook
The increased investment by the
Chinese government and multinational
pharmaceutical companies, as well as
other improvements (discussed below),
are creating a stronger environment for
innovative drug R&D in China. First, in the
past two decades, the Chinese regulatory
system has undergone a systematic
transformation to adapt to the emergence of
more INDs. The first drug administration
law in China was enacted in 1985, and
there have been four major amendments
since, with the latest one enacted in 2007.
As mentioned above, the SFDA (which
itself was founded in 2003) has introduced
more robust regulations regarding new drug
approval and registration, and to improve
transparency and efficiency, which could
pave the way for the emergence of more
innovative drugs in the long term. For
example, the registration status of an IND
or NDA is publicly available, and applicants
have easy access to all information regarding
the approval process for their drugs. Local
agencies of the SFDA are authorized to
conduct preliminary approval procedures
to increase efficiency. Companies that
provide false information or samples will be
penalized and barred from submitting NDAs
for up to 3 years.
Patent protection is a second factor that
is essential in innovative drug development.
The current Chinese patent law was enacted
NEWS & ANALYSI S
NATURE REVIEWS | DRUG DISCOVERY VOLUME 10 | MAY 2011 | 333
nrd_3435_may11.indd 333 14/04/2011 11:49
pharmacodynamic analyses. Basic Clin. Pharmacol.
Toxicol. 96, 225227 (2005).
46. Urquhart, J. Getting a handle on why good drugs
sometimes dont work. J.R.Coll. Physicians Edinb.
34, 9598 (2004).
47. Centers for Disease Control and Prevention.
Achievements in Public Health, 19001999 Family
Planning. MMWR Morb. Mortal. Wkly. Rep. 48,
10731080 (1999).
48. Vrijens, B. et al. Modellling the association between
adherence and viral load in HIV-infected patients. Stat.
Med. 24, 27192731 (2005).
49. Vrijens, B. & Urquhart, J. Patient adherence to
prescribed antimicrobial drug dosing regimens.
J.Antimicrob. Chemother. 55, 616627 (2005).
50. Olfson, M., West, J. C., Wilk, J. E. & Marcus, S. Factors
affecting the effectiveness of clinical decisions in
treating schizophrenia. in Proc. of the American
Psychiatric Assoc. 156th Annual Meeting
(1722May 2003; San Francisco, California, USA;
Abstract S28C).
51. Vrijens, B., Vincze, G., Kristanto, P., Urquhart, J. &
Burnier, M. Adherence to prescribed antihypertensive
drug treatments: longitudinal study of electronically
compiled dosing histories. BMJ 336, 11141117
(2008).
52. Vincent, O. et al. Effect of concomitant CYP2D6
inhibitor use and tamoxifen adherence on breast
cancer recurrence in early-stage breast cancer. J. Clin.
Oncol. 28, 24232429 (2010).
53. Saevarsdottir, S. et al. Patients with early rheumatoid
arthritis who smoke are less likely to respond to
treatment with methotrexate and tumor necrosis
factor inhibitors. Arthtitis Rheum. 63. 2636 (2011).
54. Harter, J. G. & Peck, C. C. Chronobiology: suggestions
for integrating it into drug development. Ann.
N.Y.Acad. Sci. 618, 563571 (1991).
55. Rothwell, P. M. Factors that can affect the external
validity of randomised controlled trials. PLoS Clin.
Trials 1, e9 (2006).
56. Thorpe, K. E. et al. A pragmatic-explanatory
continuum indicator summary (PRECIS): a tool to help
trial designers. CMAJ 180, e47e57 (2009).
57. Temple, R. Enrichment of clinical study populations.
58. Goldstein, J. Why Medicare Pays so Much For Cancer
Drugs. European Medicines Agency. Wall Street
Journal [online], http://blogs.wsj.com/
health/2009/01/27/why-medicare-pays-so-much-for-ca
ncer-drugs/tab/comments/ (2009).
59. Barbui, C. & Garattini, S. Regulatory policies on
medicines for psychiatric disorders: is Europe on
target? Br. J.Psychiatry 190, 9193 (2007).
60. Woosley, R. L. & Rice, G. A new system for moving drugs
to market. Issues Sci. Technol. XXI, 6368 (2005).
61. Eichler, H. G., Pignatti, F., Flamion, B., Leufkens, H. &
Breckenridge, A. Balancing early market access to new
drugs with the need for benefit/risk data: a mounting
dilemma. Nature Rev. Drug Discov. 7, 818826
(2008).
62. [No authors listed.] Road map to 2015. European
docs/en_GB/document_library/Report/2011/01/
WC500101373.pdf (2010).
63. Maxwell, S., Eichler, H. G., Bucsics, A., Haefeli, W. E. &
Gustafsson, L.L. on behalf of the e-SPC consortium.
e-SPC delivering drug information in the 21st
century: developing new approaches to deliver drug
information to prescribers. Br. J.Clin. Pharmacol. 6Apr
2011 (doi:10.1111/j.136521252011.03981.x.).
64. Urquhart, J. Patient noncompliance with drug
regimens: measurement, clinical correlates, economic
impact. Eur. Heart J. 17 (Suppl. A), 815 (1996).
65. Leufkens, H. G. & Urquhart, J. Variability in patterns of
drug usage. J.Pharm. Pharmacol. 46 (Suppl. 1),
433437 (1994).
66. McQuay, H. J. & Moore, R. A. Using numerical results
from systematic reviews in clinical practice. Ann.
Intern. Med. 126, 712720 (1997).
67. Woodcock J. Assessing the clinical utility of diagnostics
used in drug therapy. Clin. Pharmacol. Ther. 88,
765773 (2010).
68. Mok, T. S. et al. Gefitinib or carboplatin-paclitaxel in
pulmonary adenocarcinoma. N.Engl. J. Med. 361,
947957 (2009).
69. Donnelly, L. A. et al. A paucimorphic variant in the
HMG-CoA reductase gene is associated with lipid-
lowering response to statin treatment in diabetes: a
GoDARTS study. Pharmacogenet. Genomics 18,
10211026 (2008).
70. Flockhart, D. A. Pharmacogenetic testing of CYP2C9
and VKORC1 alleles for warfarin. Genet. Med. 10,
139150 (2008).
71. Ginsburg, G. S. & Voora, D. The long and winding road
to warfarin pharmacogenetic testing. J.Am. Coll.
Cardiol. 55, 28132815 (2010).
72. Kamali, F. & Wynne, H. Pharmacogenetics of warfarin.
Annu. Rev. Med. 61, 6375 (2010).
73. Schilsky, R. L. Personalised medicine in oncology.
Drug Discov. 7, 807817 (2008).
75. European Medicines Agency. Reflection paper on
co-development of pharmacogenomic biomarkers and
assays in the context of drug development. European
docs/en_GB/document_library/Scientific_
guideline/2010/07/WC500094445.pdf (2010).
76. FDA. International Pharmaceutical Regulatory
Monitor: US FDA draft guidance explains new drug
development tools. FDA News [online], http://www.
fdanews.com/newsletter/article?issueId=14226&artic
leId=131972 (2010).
77. European Medicines Agency. Qualification of novel
methodologies for drug development: guidance to
applicants. European Medicines Agency [online],
http://www.ema.europa.eu/docs/en_GB/document_
library/Regulatory_and_procedural_
guideline/2009/10/WC500004201.pdf (2009).
78. McClellan, M. et al. An accelerated pathway for
targeted cancer therapies. Nature Rev. Drug Discov.
10, 7980 (2011).
79. [No authors listed.] Rethinking therapeutic cancer
vaccines. Nature Rev. Drug Discov. 8, 685686
(2009).
80. Hampel, H. et al. Biomarkers for Alzheimers disease
therapeutic trials. Prog. Neurobiol. 2Dec 2010
(doi:10.1016/j.pneurobio.2010.11.005).
81. Falissard, B. et al. Relative effectiveness assessment of
listed drugs (REAL): a new method for an early
comparison of the effectiveness of approved health
technologies. Int. J.Technol. Assess Health Care 26,
124130 (2010).
82. Birkett, D. et al. Clinical pharmacology in research,
teaching and health care: considerations by IUPHAR,
the Inter-national Union of Basic and Clinical
Pharmacology. Basic Clin. Pharmacol. Toxicol. 107,
531559 (2010).
83. Pocock, S. J., Assmann, S. E., Enos, L. E. & Kasten, L. E.
Subgroup analysis, covariate adjustment and
baseline comparisons in clinical trial reporting: current
practice and problems. Stat. Med. 21, 29172930
(2002).
84. Aarons, L. et al. Role of modelling and simulation in
Phase I drug development. Eur. J.Pharm. Sci. 13,
115122 (2001).
85. Manolis, E. & Pons, G. Proposals for model-based
paediatric medicinal development within the current
European Union regulatory framework. Br. J.Clin.
Pharmacol. 68, 493501 (2009).
86. [No authors listed.] How to reduce prescribing errors.
Lancet 374, 1945 (2009).
87. [No authors listed.] Evidence-based Practice Centres:
synthesizing scientific evidence to improve quality and
effectiveness in health care. Agency for Healthcare
Research and Quality [online], http://www.ahrq.gov/
clinic/epc/ (2011).
88. Bahri, P. Public pharmacovigilance communication:
a process calling for evidence-based, objective-driven
strategies. Drug Saf. 33,10651079 (2010).
89. Smalley, W. et al. Contraindicated use of cisapride:
impact of food and drug administration regulatory
action. JAMA 284, 30363039 (2000).
90. John, J. et al. HER2 testing and subsequent
trastuzumab treatment for breast cancer in a
managed care environment. Oncologist 14, 760768
(2009).
91. Gustafsson, L. L. et al. The Wise List a
comprehensive concept to select, communicate and
achieve recommendations of essential drugs in
ambulatory care in Stockholm. Basic Clin. Pharmacol.
Toxicol.108, 224233 (2011).
92. Shea, S. & Hripcsak, G. Accelerating the use of
electronic health records in physician practices.
N.Engl. J.Med. 362, 192195 (2010).
93. Sjborg, B. et al. Design and implementation of a
point-of-care computerized system for drug therapy in
Stockholm metropolitan health region-bridging the
gap between knowledge and practice. Int. J.Med.
Inform. 76, 497506 (2007).
94. Classen, D. C., Avery, A. J. & Bates, D. W. Evaluation
and certification of computerized provider order entry
systems. J.Am. Med. Inform. Assoc. 14, 4855
(2007).
95. Bttiger, Y. et al. SFINX a drugdrug interaction
database desgined for clinical decision support
systems. Eur. J.Clin. Pharmacol. 95, 627633
(2009).
96. Wettermark, B. et al. The new Swedish Prescribed
Drug Registeropportunities for
pharmacoepidemiological research and experience
from the first six months. Pharmacoepidemiol. Drug
Saf. 16, 726735 (2007).
97. Cutler, D. & Everett, W. Thinking outside the pillbox
medication adherence as a priority for health care
reform. N.Engl. J.Med. 362, 15531555
(2010).
98. Modi, A. C. & Quittner, A. L. Barriers to treatment
adherence for children with cystic fibrosis and asthma:
what gets in the way? J.Pediatr. Psychol. 31,
846858 (2006).
99. Osterberg, L. G., Urquhart, J. & Blaschke, T. F.
Understanding forgiveness: minding and mining the
gaps between pharmacokinetics and therapeutics.
100. Frieden, T. R., Fujiwara, P. I., Washko, R. M. &
Hamburg, M. A. Tuberculosis in New York City
turning the tide. N.Engl. J.Med. 333, 229233
(1995).
101. Lester, R. T. et al. Effects of a mobile phone short
message service on antiretroviral treatment adherence
in Kenya (WelTel Kenya1): a randomised trial. Lancet
376, 18381845 (2010).
102. Chi, B. H. & Stringer, J. S. A. Mobile phones to
improve HIV treatment adherence. Lancet 376,
18071808 (2010).
103. Ogedegbe G. Self-titration for treatment of
uncomplicated hypertension. Lancet 376, 144146
(2010).
104. Singer, E. Message from a pill bottle. Technology
Review [online], http://www.technologyreview.com/
business/26414/?p1=A4 (2010).
105. Au-Yeung, K. Y. et al. A networked system for self-
management of drug therapy and wellness. ACM
Digital Library [online], http://delivery.acm.
org/10.1145/1930000/1921083/p1-au-yeung.pdf?ke
y1=1921083&key2=9929465921&coll=DL&dl=
ACM&CFID=7014173&CFTOKEN=96965669
(2010).
106. Sokol, M. C., McGuigan, K. A., Verbrugge, R. R. &
Epstein, R. S. Impact of medication adherence on
hospitalization risk and healthcare cost. Med. Care
43, 521530 (2005).
107. Ware, J. H. & Hamel, M. B. Pragmatic trials guide
to better patient care? N.Engl. J. Med. 364,
16851687 (2011).
108. Hughes, B. 2009 FDA drug approvals. Nature Rev.
Drug Discov. 9, 8992 (2010).
109. Eichler, H. G., Aronsson, B., Abadie, E. & Salmonson, T.
New drug approval success rate in Europe
in 2009. Nature Rev. Drug Discov. 9, 355356
(2010).
PERSPECTI VES
nrd_3501_jul11.indd 506 20/06/2011 16:34
First published in Nature Reviews Drug Discovery 10, 333-334 (2011); doi:10.1038/nrd3435
The impact of mergers on
pharmaceutical R&D
John L.LaMattina
Mergers and acquisitions in the pharmaceutical industry have substantially reduced the
number of major companies over the past 15years. The short-term business rationale for this
extensive consolidation might have been reasonable, but at what cost to research and
development productivity?
John L.LaMattina is the
former President of Pfizer
Global Research and
Development, and is currently
Senior Partner at Puretech
Ventures, Boston,
e-mail: john.lamattina@
comcast.net
doi:10.1038/nrd3514
Concerns about the productivity of pharmaceutical
research and development (R&D) are becoming increas-
ingly common in both the mainstream media and sci-
entific literature. A range of possible causes have been
identified, from more challenging therapeutic targets
to excessive bureaucracy, and various approaches to
address these issues have been put forward (for example,
see REFS 1,2). However, the impact of mergers and acqui-
sitions on R&D productivity is less well documented,
because R&D integrations and cuts are largely done pri-
vately. In this article, it is argued that although mergers
and acquisitions in the pharmaceutical industry might
have had a reasonable short-term business rationale,
their impact on the R&D of the organizations involved
has been devastating.
Industry consolidation
When people bemoan the poor productivity of the phar-
maceutical industry at present, they often refer back to
the heyday of new drug approvals by the US Food and
Drug Administration (FDA): the 1990s. Indeed, in terms
of the number of new drugs that were approved, this
decade was more productive, with an average of 31 drugs
per year between 1990 and 1999 (compared with 24 per
year between 2000 and 2009), with a peak of 54 drugs in
1996. One possible contributory factor is that multiple
entries in a single drug class (such as statins) were more
economically viable at thetime.
However, another underlying factor contributing
to the productivity observed in the 1990s was the large
number of pharmaceutical companies at that time.
Many of the drugs that were approved in1996 origi-
nated from companies that no longer exist; indeed, out
of the 42 members of the Pharmaceutical Research and
Manufacturers of America (PhRMA) in 1988, only 11
(~25%) remain today (see Supplementary information
S1 (figure)).
The R&D portfolios of these companies, although
differing in size, tended to be broader in scope than
those of start-up companies that arose during this time,
and it is likely that when a new idea for treating cancer
arose in 1990, 20 companies would have initiated pro-
jects on it. Given the difficulties that are encountered
in R&D, it could reasonably be assumed that only three
or four of these companies would have been success-
ful at bringing a drug based on this idea to market.
Furthermore, the greater diversity of portfolios among
a larger number of companies both large and small
could increase the chances of finding new drugs in
general. Indeed, a recent analysis has indicated that the
number of new drug approvals during the past 60years
is correlated with the number of companies
3
. Now, with
so many fewer major companies involved in pharmaceu-
tical R&D, the chances of success in the industry overall
are likely to be dropping precipitously.
R&D reductions
From a business perspective, mergers and acquisitions
are often considered to be attractive as they remove
duplication, reduce costs and produce synergies.
Furthermore, in the early days of mergers in the phar-
maceutical industry, organizations often described them
as being part of a growth story. In these situations
for example, the merger of Bristol Myers with Squibb
in 1989 the R&D divisions were fused. Programme
overlap was minimized and new projects were added,
and major R&D cuts did notoccur.
This has changed radically in the past decade. In
major mergers today, not only are R&D cuts made, but
entire research sites are eliminated. Nowhere is this
more evident than with Pfizer. Before 1999, Pfizer had
never made a major acquisition. Over the next decade,
it acquired three large companies Warner-Lambert
(in 2000), Pharmacia (in 2003) and Wyeth (in 2009)
and multiple smaller companies, such as Vicuron, Rinat
and Esperion (Supplementary information S1 (figure)).
Over this time frame, to meet its business objectives
(a euphemism for raising its stock price) Pfizer closed
COMMENT
NATURE REVIEWS | DRUG DISCOVERY VOLUME 10 | AUGUST 2011 | 559
nrd_3514_aug11.indd 559 19/07/2011 15:03
N
u
m
b
e
r

o
f

d
r
u
g
s
0
20
40
60
80
100
120
I II III
C
Phase Region
10
20
30
40
50
60
70
80
N
u
m
b
e
r

o
f

d
r
u
g
s
0
CN US EU
E
110
10
23
16
36
41
Cancer
Infection
Cardiovascular
Central nervous system
Endocrine/metabolism
Musculoskeletal
Respiratory
Genitourinal
Gastrointestinal
Others
D
32
11
10
1
1
S
13
3
3
3
R&D IN CHINA | MARKET INDICATORS
in 1984, but until it was amended in 1992,

pharmaceutical compositions were not
patentable. Now, the patent system has
evolved to provide greater protection for
innovative drugs. The final judgment from
the Beijing High Peoples Court in 2007
on the patent dispute over sildenafil
(Viagra; Pfizer) is one example. The
judgment rejected the patent challenge from
12 domestic generics companies, effectively
providing patent protection for sildenafil
until 2014. Interestingly, the patent that
was under dispute is a method-of-use
patent, which is more vulnerable to
challenges from generics companies than
compound patents.
Third, to improve investment, ChiNext,
Chinas NASDAQ, was launched in
2009, focusing on innovative enterprises
and other fledgling venture enterprises.
ChiNext provides an important exit for
investment, such as venture capital in the
field of innovative R&D. Fourth, China
initiated a health-care reform plan in 2009,
which will create a demand for innovative
pharmaceutical products in the years
to come. Finally, with regard to talent,
the current wave of returnees to China
includes many experienced professionals
from pharmaceutical and biotechnology
companies elsewhere in the world.
In conclusion, although the innovative
pharmaceutical industry in China is still in the
early stages of development, it has progressed
rapidly in the past decade. Moreover, China
has made considerable improvements and
continues to improve the key factors
regulatory systems, patent protection, basic
research, investment, talent and market
incentives for the creation of a first-class
environment for innovative drug R&D.
Jingzong Qi and Xin Chen are at FusoGen
Pharmaceuticals,19F-A, Ping An Building B, 59
Machang Road, Hexi District, Tianjin 300203, China.
Qingli Wang is at the Center for Drug Evaluation, State
Food and Drug Administration, Beijing 100038, China.
Zhenhang Yu is at the China National Center for
Biotechnology Development, Beijing 100039, China.
Fengshan Wang is at the Institute of Biochemical and
Biotechnological Drugs, School of Pharmaceutical
Sciences, Shandong University, Jinan 250012,
Shandong, China.
Correspondence to F.W. e-mail: fswang@sdu.edu.cn
1. IMS Intelligence Applied. IMS Health forecasts global
pharmaceutical market growth of 5-7 percent in 2011,
reaching $880 billion. IMS website [online], http://www.
imshealth.com/portal/site/imshealth/menuitem.a46c6d
4df3db4b3d88f611019418c22a/?vgnextoid=119717
f27128b210VgnVCM100000ed152ca2RCRD&cpsextc
urrchannel=1 (2010).
Competing financial interests
Table 1 | Novel drugs in clinical trials with issued compound patents in the US or EU
Compound Developer Therapeutic area Patent Stage
Iptakalim Nhwa Cardiovascular EU III
Aildenafil Wannianchun Genitourinal US III
Cymipristone Xianju Genitourinal US/EU III
Icotinib Beida Cancer US III
rSIFN Huiyang Cancer US/EU III
Sulcardine SIMM Cardiovascular US/EU II
Trichosanthin Xiangtianmu Infection US II
Sifuvirtide FusoGen Infection US/EU II
Chiglitazar Chipscreen Endocrine/metabolism US II
Tyroservatide Yitai Cancer US II
Trantinterol Jiutai Respiratory US/EU II
Chidamide Chipscreen Cancer US II
Buagafuran SIMM Central nervous system US I
Ethaselen Qizheng Cancer US I
Nemonoxacin Huayu Infection US/EU I
Albuvirtide Frontier Infection US I
Thienorphine IPT Central nervous system US/EU I
Triptolide SIMM Musculoskeletal US I
Novaferon Genova Cancer US I
PEG-rhArginase BCT Cancer EU I
yPEGrhGCSF Amoytop Cancer EU I
yPEGrhGH Amoytop Endocrine/metabolism EU I
Citricplatin Sansiweier Cancer US/EU I
yPEGrhIFN2a Amoytop Infection EU I
EU, European Union; GCSF, granulocyte colony-stimulating factor; GH, growth hormone; IPT, Institute of
Pharmacology and Toxicology; PEG, polyethylene glycol; rh, recombinant human; rSIFN, recombinant super
compound interferon (IFN); SIMM, Shanghai Institute of Materia Medica; SIPI, Shanghai Institute of
Pharmaceutical Industry; US, United States of America; yPEG, Y-shape branched PEG.
Figure 2 | Characteristics of novel
investigational drugs in China. a | Number
of drugs in each phase of clinical trials.
b | Indications of these drugs. c | Patent
protection of these drugs. Some drugs have
patent protection not only in China (CN), but
also in the United States (US) or the European
Union (EU). Data were collected by the end of
2010 from websites of the Chinese State Food
and Drug Administration (SFDA) and the Center
for Drug Evaluation of the SFDA, the Chinese
State Intellectual Property Office, the US Patent
and Trademark Office, the European Patent
Office and company press releases.
NEWS & ANALYSI S
334 | MAY 2011 | VOLUME 10 www.nature.com/reviews/jrnl
nrd_3435_may11.indd 334 14/04/2011 11:49
The impact of mergers on
pharmaceutical R&D
John L.LaMattina
Mergers and acquisitions in the pharmaceutical industry have substantially reduced the
number of major companies over the past 15years. The short-term business rationale for this
extensive consolidation might have been reasonable, but at what cost to research and
development productivity?
John L.LaMattina is the
former President of Pfizer
Global Research and
Development, and is currently
Senior Partner at Puretech
Ventures, Boston,
e-mail: john.lamattina@
comcast.net
doi:10.1038/nrd3514
Concerns about the productivity of pharmaceutical
research and development (R&D) are becoming increas-
ingly common in both the mainstream media and sci-
entific literature. A range of possible causes have been
identified, from more challenging therapeutic targets
to excessive bureaucracy, and various approaches to
address these issues have been put forward (for example,
see REFS 1,2). However, the impact of mergers and acqui-
sitions on R&D productivity is less well documented,
because R&D integrations and cuts are largely done pri-
vately. In this article, it is argued that although mergers
and acquisitions in the pharmaceutical industry might
have had a reasonable short-term business rationale,
their impact on the R&D of the organizations involved
has been devastating.
Industry consolidation
When people bemoan the poor productivity of the phar-
maceutical industry at present, they often refer back to
the heyday of new drug approvals by the US Food and
Drug Administration (FDA): the 1990s. Indeed, in terms
of the number of new drugs that were approved, this
decade was more productive, with an average of 31 drugs
per year between 1990 and 1999 (compared with 24 per
year between 2000 and 2009), with a peak of 54 drugs in
1996. One possible contributory factor is that multiple
entries in a single drug class (such as statins) were more
economically viable at thetime.
However, another underlying factor contributing
to the productivity observed in the 1990s was the large
number of pharmaceutical companies at that time.
Many of the drugs that were approved in1996 origi-
nated from companies that no longer exist; indeed, out
of the 42 members of the Pharmaceutical Research and
Manufacturers of America (PhRMA) in 1988, only 11
(~25%) remain today (see Supplementary information
S1 (figure)).
The R&D portfolios of these companies, although
differing in size, tended to be broader in scope than
those of start-up companies that arose during this time,
and it is likely that when a new idea for treating cancer
arose in 1990, 20 companies would have initiated pro-
jects on it. Given the difficulties that are encountered
in R&D, it could reasonably be assumed that only three
or four of these companies would have been success-
ful at bringing a drug based on this idea to market.
Furthermore, the greater diversity of portfolios among
a larger number of companies both large and small
could increase the chances of finding new drugs in
general. Indeed, a recent analysis has indicated that the
number of new drug approvals during the past 60years
is correlated with the number of companies
3
. Now, with
so many fewer major companies involved in pharmaceu-
tical R&D, the chances of success in the industry overall
are likely to be dropping precipitously.
R&D reductions
From a business perspective, mergers and acquisitions
are often considered to be attractive as they remove
duplication, reduce costs and produce synergies.
Furthermore, in the early days of mergers in the phar-
maceutical industry, organizations often described them
as being part of a growth story. In these situations
for example, the merger of Bristol Myers with Squibb
in 1989 the R&D divisions were fused. Programme
overlap was minimized and new projects were added,
and major R&D cuts did notoccur.
This has changed radically in the past decade. In
major mergers today, not only are R&D cuts made, but
entire research sites are eliminated. Nowhere is this
more evident than with Pfizer. Before 1999, Pfizer had
never made a major acquisition. Over the next decade,
it acquired three large companies Warner-Lambert
(in 2000), Pharmacia (in 2003) and Wyeth (in 2009)
and multiple smaller companies, such as Vicuron, Rinat
and Esperion (Supplementary information S1 (figure)).
Over this time frame, to meet its business objectives
(a euphemism for raising its stock price) Pfizer closed
COMMENT
NATURE REVIEWS | DRUG DISCOVERY VOLUME 10 | AUGUST 2011 | 559
nrd_3514_aug11.indd 559 19/07/2011 15:03
N
u
m
b
e
r

o
f

d
r
u
g
s
0
20
40
60
80
100
120
I II III
C
Phase Region
10
20
30
40
50
60
70
80
N
u
m
b
e
r

o
f

d
r
u
g
s
0
CN US EU
E
110
10
23
16
36
41
Cancer
Infection
Cardiovascular
Central nervous system
Endocrine/metabolism
Musculoskeletal
Respiratory
Genitourinal
Gastrointestinal
Others
D
32
11
10
1
1
S
13
3
3
3
R&D IN CHINA | MARKET INDICATORS
in 1984, but until it was amended in 1992,

pharmaceutical compositions were not
patentable. Now, the patent system has
evolved to provide greater protection for
innovative drugs. The final judgment from
the Beijing High Peoples Court in 2007
on the patent dispute over sildenafil
(Viagra; Pfizer) is one example. The
judgment rejected the patent challenge from
12 domestic generics companies, effectively
providing patent protection for sildenafil
until 2014. Interestingly, the patent that
was under dispute is a method-of-use
patent, which is more vulnerable to
challenges from generics companies than
compound patents.
Third, to improve investment, ChiNext,
Chinas NASDAQ, was launched in
2009, focusing on innovative enterprises
and other fledgling venture enterprises.
ChiNext provides an important exit for
investment, such as venture capital in the
field of innovative R&D. Fourth, China
initiated a health-care reform plan in 2009,
which will create a demand for innovative
pharmaceutical products in the years
to come. Finally, with regard to talent,
the current wave of returnees to China
includes many experienced professionals
from pharmaceutical and biotechnology
companies elsewhere in the world.
In conclusion, although the innovative
pharmaceutical industry in China is still in the
early stages of development, it has progressed
rapidly in the past decade. Moreover, China
has made considerable improvements and
continues to improve the key factors
regulatory systems, patent protection, basic
research, investment, talent and market
incentives for the creation of a first-class
environment for innovative drug R&D.
Jingzong Qi and Xin Chen are at FusoGen
Pharmaceuticals,19F-A, Ping An Building B, 59
Machang Road, Hexi District, Tianjin 300203, China.
Qingli Wang is at the Center for Drug Evaluation, State
Food and Drug Administration, Beijing 100038, China.
Zhenhang Yu is at the China National Center for
Biotechnology Development, Beijing 100039, China.
Fengshan Wang is at the Institute of Biochemical and
Biotechnological Drugs, School of Pharmaceutical
Sciences, Shandong University, Jinan 250012,
Shandong, China.
Correspondence to F.W. e-mail: fswang@sdu.edu.cn
1. IMS Intelligence Applied. IMS Health forecasts global
pharmaceutical market growth of 5-7 percent in 2011,
reaching $880 billion. IMS website [online], http://www.
imshealth.com/portal/site/imshealth/menuitem.a46c6d
4df3db4b3d88f611019418c22a/?vgnextoid=119717
f27128b210VgnVCM100000ed152ca2RCRD&cpsextc
urrchannel=1 (2010).
Table 1 | Novel drugs in clinical trials with issued compound patents in the US or EU
Compound Developer Therapeutic area Patent Stage
Iptakalim Nhwa Cardiovascular EU III
Aildenafil Wannianchun Genitourinal US III
Cymipristone Xianju Genitourinal US/EU III
Icotinib Beida Cancer US III
rSIFN Huiyang Cancer US/EU III
Sulcardine SIMM Cardiovascular US/EU II
Trichosanthin Xiangtianmu Infection US II
Sifuvirtide FusoGen Infection US/EU II
Chiglitazar Chipscreen Endocrine/metabolism US II
Tyroservatide Yitai Cancer US II
Trantinterol Jiutai Respiratory US/EU II
Chidamide Chipscreen Cancer US II
Buagafuran SIMM Central nervous system US I
Ethaselen Qizheng Cancer US I
Nemonoxacin Huayu Infection US/EU I
Albuvirtide Frontier Infection US I
Thienorphine IPT Central nervous system US/EU I
Triptolide SIMM Musculoskeletal US I
Novaferon Genova Cancer US I
PEG-rhArginase BCT Cancer EU I
yPEGrhGCSF Amoytop Cancer EU I
yPEGrhGH Amoytop Endocrine/metabolism EU I
Citricplatin Sansiweier Cancer US/EU I
yPEGrhIFN2a Amoytop Infection EU I
EU, European Union; GCSF, granulocyte colony-stimulating factor; GH, growth hormone; IPT, Institute of
Pharmacology and Toxicology; PEG, polyethylene glycol; rh, recombinant human; rSIFN, recombinant super
compound interferon (IFN); SIMM, Shanghai Institute of Materia Medica; SIPI, Shanghai Institute of
Pharmaceutical Industry; US, United States of America; yPEG, Y-shape branched PEG.
Figure 2 | Characteristics of novel
investigational drugs in China. a | Number
of drugs in each phase of clinical trials.
b | Indications of these drugs. c | Patent
protection of these drugs. Some drugs have
patent protection not only in China (CN), but
also in the United States (US) or the European
Union (EU). Data were collected by the end of
2010 from websites of the Chinese State Food
and Drug Administration (SFDA) and the Center
for Drug Evaluation of the SFDA, the Chinese
State Intellectual Property Office, the US Patent
and Trademark Office, the European Patent
Office and company press releases.
NEWS & ANALYSI S
334 | MAY 2011 | VOLUME 10 www.nature.com/reviews/jrnl
nrd_3435_may11.indd 334 14/04/2011 11:49
First published in Nature Reviews Drug Discovery 10, 559-560 (August 2011); doi: 10.1038/nrd3514
Crowd sourcing in drug discovery
Crowd sourcing is emerging as an open-innovation approach to promote collaboration and
harness the complementary expertise of academic and industrial partners in the early stages of
drug discovery. Here, we highlight examples of such initiatives and discuss key success factors.
Greater collaboration between academic institutions and
the pharmaceutical industry is increasingly being pur-
sued to access and foster innovation in the early stages
of drug discovery. The hope is that such collaborations
could help to address the need to improve research and
development (R&D) productivity in industry
1
, and also
enable academic institutions to more effectively exploit
the translational potential of their research. Although
there are a growing number of examples of industry
collaborating closely with a major academic partner,
approaches that harness the expertise of a larger sci-
entific community to address a specific question have
been more limited. Nevertheless, using the internet as
a platform, an open-innovation model known as crowd
sourcing
2
is now being tested in early-stage drug dis-
covery by several organizations. This model has suc-
cessfully been used in other sectors (for example, the
Procter & Gamble connect and develop portal (www.
pgconnectdevelop.com) through which consumers can
put forward their ideas for product improvements or
novel products).
Applications of crowd sourcing
Originally, crowd sourcing was defined as a mechanism
by which specific problems are communicated to an
unknown group of potential solvers in the form of an
open call, usually via the internet; the community (the
crowd) is asked to provide solutions and the winners
are rewarded. In 2001, Eli Lilly was the first company to
introduce this concept in drug discovery, with the estab-
lishment of the InnoCentive platform (www.innocen-
tive.com). Organizations in need of answers (seekers)
post specific questions (challenges) on an internet mar-
ketplace. The web community can then provide solu-
tions to the challenge (solvers). In each challenge, the
seeking company can select the best solution and the
winning solver transfers the intellectual property (IP) to
the seeker in return for a financial reward. InnoCentive
is now an independent organization with a solver com-
munity of more than 200,000 experts from more than
20 countries.
Further crowd sourcing initiatives in the area of early-
stage drug discovery have followed in the past 2 years
(see Supplementary information S1 (table) for more
details of the examples discussed below). Importantly,
in contrast to the classical concept of crowd sourcing,
in which the task is finished once the solution has been
provided, the goal of these initiatives is to seek novel
ideas that are then pursued further in a more collabora-
tive approach. The key benefits are that potential solvers,
most of whom are researchers in academic institutes or
small companies, gain access to specific tools or knowl-
edge, such as data, assays, compounds or drug discovery
expertise in large pharmaceutical companies, whereas
the searching organization gains novel ideas, targets,
compounds or tools (such as novel assays or models)
that help to address a specific challenge.
For example, in May 2009 Bayer Healthcare intro-
duced its Grants4Targets initiative (G4T; www.grant-
s4targets.com)
3
, the goal of which is to discover new
therapeutic options by bringing together knowledge
on potential novel targets in academia with drug devel-
opment expertise within the company. Grants for the
validation of innovative targets in oncology, cardiology,
molecular imaging and gynaecology are provided for a
period of 1 year. To promote the validation process and
generate value for both partners, senior scientists from
the company are appointed to support the projects with
expertise and tools. Three types of grants are provided
(support, focus and collaborative), and IP remains fully
with the applicants for the support and the focus grants
3
.
After the grant period, promising targets may be pursued
further via collaborative agreements. So far, four calls
have been completed, more than 380 grant applications
have been received, and 59 grants have been awarded to
academic groups worldwide
3
.
A similar approach was recently initiated by MRC
Technology (MRCT), the technology transfer arm of the
Medical Research Council (MRC) UK. MRCTs Centre
for Therapeutics Discovery is focused on de-risking
early-stage academic projects in areas of substantial
unmet medical need, showing their potential in pre-
clinical models before partnering with industry. This
resource was originally developed to translate MRC-
based research, but in 2010 they started their Call for
Targets programme (www.callfortargets.org) to identify
early-stage projects from non-MRC sources. The review
process has two stages: an initial triage, which is followed
Monika Lessl and Khusru
Asadullah are at Global Drug
Discovery Bayer HealthCare
Pharmaceuticals,
Muellerstrae 178, 13342
Berlin, Germany.
Justin. S.Bryans is at the
Centre for Therapeutics
Discovery, MRC Technology,
13 Burtonhole Lane, London
NW7 1AD, UK.
Duncan Richards is at
GlaxoSmithKIine, Academic
Discovery Performance Unit,
Medicines Research Centre,
Stevenage SG1 2NY, UK.
Correspondence to M.L.
e-mail: Monika.Lessl@bayer.
com
COMMENT
NATURE REVIEWS | DRUG DISCOVERY VOLUME 10 | APRIL 2011 | 241
nrd_3412_apr11.indd 241 17/03/2011 11:40
numerous research sites in the United States, includ-
ing those at Kalamazoo, Michigan (formerly a site for
Upjohn), Ann Arbor, Michigan (formerly a site for
Warner-Lambert) and Skokie, Illinois (formerly a site
for Searle). It has also recently announced the closure of
the Sandwich site in the UK. These sites housed thou-
sands of scientists, and many major drugs such as
atorvastatin (Lipitor), amlodipine (Norvasc) and silde-
nafil (Viagra) were discovered there. The same pat-
tern has been observed after most of the mergers and
acquisitions by other major pharmaceutical companies
during the pastdecade.
There is another key aspect to such cuts. Historically,
the pharmaceutical industry has prided itself on invest-
ing more in R&D (as a percentage of revenues) than any
other industry. At times, companies have invested as
much as 20% of top-line revenues into their pipeline.
However, Pfizer now projects that in 2012 this figure
will only be 11% (between US$6.5 and $7billion). The
extent of this decrease is further emphasized by com-
parison with the pre-merger R&D expenses of Pfizer and
Wyeth in 2008: $7.95billion and $3.37billion, respec-
tively, and $11.3billion in total. Other large pharma-
ceutical companies have announced similar cuts in
recentyears.
Thus, at a time when our understanding of the basis
of diseases continues to increase substantially, the ability
to exploit this information in the private sector is being
compromised. It is hard to envision that R&D output,
as measured by new drug approvals, will improve in the
coming years based on this reduced investment.
Pipeline advancement
After a major merger, the rate of progress of compounds
in the development pipeline seems to decrease. For
example, comparing data from Pfizers pipeline updates
(which are posted on its website every 6 months) before
the Wyeth merger in February 2008, and in February
2011, reveals that 40% of the compounds (not including
those from Wyeth) have been in Phase II development
for more than 3 years, which is below the industry aver-
age (J. Arrowsmith, personal communication).
Indeed, R&D seems to be especially vulnerable to
the negative impact of mergers and acquisitions. Having
a sense of how mergers occur in R&D organizations
is helpful for understanding this impact. R&D organi-
zations will be the last part of the companies to begin
merger discussions before regulatory approval because
of the commercial sensitivity of the pipeline and the
intellectual property of the company. And when the
discussions about integrating the R&D organiza-
tions finally occur, the initial focus is on Phase III
programmes, followed by mid-stage candidates, with
the early-stage discovery programmes handled last.
These reviews are extensive and time-consuming, as
they require careful consideration of scientific issues
such as efficacy and safety data for each programme,
as well as commercial issues such as potential duplica-
tion and strategic directions of the merged company. In
addition, research organizations often differ proce-
durally in some fundamental processes such as IT
platforms, data handling or adverse event monitoring.
Establishing which system to use or creating a hybrid
takes substantial time for decision-making as well as
implementation.
It is easy to see how early-stage R&D will be slowed
in such situations, as during this period which can
take at least 9months generally no new programmes
are started and hiring will be frozen. Undergoing one
merger will have a substantial negative impact on the
momentum of research programmes, but enduring this
multiple times can be crippling.
Social consequences
For individual employees of companies that are in the
process of merging, uncertainty about their future is
often high. It is hard to quantify the impact that such
uncertainty has on productivity, except that it is nega-
tive, and probably strongly so in many cases. Leaders
of organizations who have completed multiple mergers
may express the view: Weve done this before and we
know how to do it. Mechanically, this may be true, but
for many employees who survive mergers, the thought of
repeating the exercise is not embraced, and could prove
to be numbing to their motivation.
Going forward
It is unlikely that the era of large mergers in the phar-
maceutical industry has ended. However, as a strategy
to achieve top-line growth, it has a clear flaw if the
R&D engine is not growing robustly enough to keep
pace, more mergers and acquisitions will be needed to
continuegrowth.
Not all CEOs of major pharmaceutical companies
believe in this strategy; for example, John Lechleiter,
CEO of Lilly, has stated his opposition to large-scale
combination. And not all CEOs of merged companies
believe in cutting back R&D; for example, Mercks new
CEO, Kenneth Frazier, has recently stated that Merck
will focus on investing in drug development to drive
growth. Whether other leaders in the pharmaceutical
industry will come around to the views of Lechleiter
and Frazier remains to be seen. However, the experience
from the past decade on the negative impact of merg-
ers and acquisitions on R&D productivity should make
these leaders pause when considering major mergers in
thefuture.
Such mergers should also be concerning to patients,
physicians and payers, particularly bearing in mind
recent cutbacks in areas of research such as antibacte-
rial drugs and neuroscience. Industry consolidation has
resulted in less competition and less investment in R&D.
At a time when there is a major need for new treatments
for conditions such as Alzheimers disease, drug-resistant
infections and diabetes, such a trend is alarming.
1. Paul, S. M. et al. How to improve R&D productivity: the
pharmaceutical industrys grand challenge. Nature Rev. Drug
Discov. 9, 203214 (2010).
2. Garnier, J.-P. Rebuilding the R&D engine in Big Pharma. Harvard
Bus. Rev. 86, 6876 (2008).
3. Munos, B. Lessons from 60years of pharmaceutical innovation.
The author declares no competing financial interests.
560 | AUGUST 2011 | VOLUME 10 www.nature.com/reviews/drugdisc
COMMENT
nrd_3514_aug11.indd 560 19/07/2011 15:03
Crowd sourcing in drug discovery
Crowd sourcing is emerging as an open-innovation approach to promote collaboration and
harness the complementary expertise of academic and industrial partners in the early stages of
drug discovery. Here, we highlight examples of such initiatives and discuss key success factors.
Greater collaboration between academic institutions and
the pharmaceutical industry is increasingly being pur-
sued to access and foster innovation in the early stages
of drug discovery. The hope is that such collaborations
could help to address the need to improve research and
development (R&D) productivity in industry
1
, and also
enable academic institutions to more effectively exploit
the translational potential of their research. Although
there are a growing number of examples of industry
collaborating closely with a major academic partner,
approaches that harness the expertise of a larger sci-
entific community to address a specific question have
been more limited. Nevertheless, using the internet as
a platform, an open-innovation model known as crowd
sourcing
2
is now being tested in early-stage drug dis-
covery by several organizations. This model has suc-
cessfully been used in other sectors (for example, the
Procter & Gamble connect and develop portal (www.
pgconnectdevelop.com) through which consumers can
put forward their ideas for product improvements or
novel products).
Applications of crowd sourcing
Originally, crowd sourcing was defined as a mechanism
by which specific problems are communicated to an
unknown group of potential solvers in the form of an
open call, usually via the internet; the community (the
crowd) is asked to provide solutions and the winners
are rewarded. In 2001, Eli Lilly was the first company to
introduce this concept in drug discovery, with the estab-
lishment of the InnoCentive platform (www.innocen-
tive.com). Organizations in need of answers (seekers)
post specific questions (challenges) on an internet mar-
ketplace. The web community can then provide solu-
tions to the challenge (solvers). In each challenge, the
seeking company can select the best solution and the
winning solver transfers the intellectual property (IP) to
the seeker in return for a financial reward. InnoCentive
is now an independent organization with a solver com-
munity of more than 200,000 experts from more than
20 countries.
Further crowd sourcing initiatives in the area of early-
stage drug discovery have followed in the past 2 years
(see Supplementary information S1 (table) for more
details of the examples discussed below). Importantly,
in contrast to the classical concept of crowd sourcing,
in which the task is finished once the solution has been
provided, the goal of these initiatives is to seek novel
ideas that are then pursued further in a more collabora-
tive approach. The key benefits are that potential solvers,
most of whom are researchers in academic institutes or
small companies, gain access to specific tools or knowl-
edge, such as data, assays, compounds or drug discovery
expertise in large pharmaceutical companies, whereas
the searching organization gains novel ideas, targets,
compounds or tools (such as novel assays or models)
that help to address a specific challenge.
For example, in May 2009 Bayer Healthcare intro-
duced its Grants4Targets initiative (G4T; www.grant-
s4targets.com)
3
, the goal of which is to discover new
therapeutic options by bringing together knowledge
on potential novel targets in academia with drug devel-
opment expertise within the company. Grants for the
validation of innovative targets in oncology, cardiology,
molecular imaging and gynaecology are provided for a
period of 1 year. To promote the validation process and
generate value for both partners, senior scientists from
the company are appointed to support the projects with
expertise and tools. Three types of grants are provided
(support, focus and collaborative), and IP remains fully
with the applicants for the support and the focus grants
3
.
After the grant period, promising targets may be pursued
further via collaborative agreements. So far, four calls
have been completed, more than 380 grant applications
have been received, and 59 grants have been awarded to
academic groups worldwide
3
.
A similar approach was recently initiated by MRC
Technology (MRCT), the technology transfer arm of the
Medical Research Council (MRC) UK. MRCTs Centre
for Therapeutics Discovery is focused on de-risking
early-stage academic projects in areas of substantial
unmet medical need, showing their potential in pre-
clinical models before partnering with industry. This
resource was originally developed to translate MRC-
based research, but in 2010 they started their Call for
Targets programme (www.callfortargets.org) to identify
early-stage projects from non-MRC sources. The review
process has two stages: an initial triage, which is followed
Monika Lessl and Khusru
Asadullah are at Global Drug
Discovery Bayer HealthCare
Pharmaceuticals,
Muellerstrae 178, 13342
Berlin, Germany.
Justin. S.Bryans is at the
Centre for Therapeutics
Discovery, MRC Technology,
13 Burtonhole Lane, London
NW7 1AD, UK.
Duncan Richards is at
GlaxoSmithKIine, Academic
Discovery Performance Unit,
Medicines Research Centre,
Stevenage SG1 2NY, UK.
Correspondence to M.L.
e-mail: Monika.Lessl@bayer.
com
COMMENT
NATURE REVIEWS | DRUG DISCOVERY VOLUME 10 | APRIL 2011 | 241
nrd_3412_apr11.indd 241 17/03/2011 11:40
numerous research sites in the United States, includ-
ing those at Kalamazoo, Michigan (formerly a site for
Upjohn), Ann Arbor, Michigan (formerly a site for
Warner-Lambert) and Skokie, Illinois (formerly a site
for Searle). It has also recently announced the closure of
the Sandwich site in the UK. These sites housed thou-
sands of scientists, and many major drugs such as
atorvastatin (Lipitor), amlodipine (Norvasc) and silde-
nafil (Viagra) were discovered there. The same pat-
tern has been observed after most of the mergers and
acquisitions by other major pharmaceutical companies
during the pastdecade.
There is another key aspect to such cuts. Historically,
the pharmaceutical industry has prided itself on invest-
ing more in R&D (as a percentage of revenues) than any
other industry. At times, companies have invested as
much as 20% of top-line revenues into their pipeline.
However, Pfizer now projects that in 2012 this figure
will only be 11% (between US$6.5 and $7billion). The
extent of this decrease is further emphasized by com-
parison with the pre-merger R&D expenses of Pfizer and
Wyeth in 2008: $7.95billion and $3.37billion, respec-
tively, and $11.3billion in total. Other large pharma-
ceutical companies have announced similar cuts in
recentyears.
Thus, at a time when our understanding of the basis
of diseases continues to increase substantially, the ability
to exploit this information in the private sector is being
compromised. It is hard to envision that R&D output,
as measured by new drug approvals, will improve in the
coming years based on this reduced investment.
Pipeline advancement
After a major merger, the rate of progress of compounds
in the development pipeline seems to decrease. For
example, comparing data from Pfizers pipeline updates
(which are posted on its website every 6 months) before
the Wyeth merger in February 2008, and in February
2011, reveals that 40% of the compounds (not including
those from Wyeth) have been in Phase II development
for more than 3 years, which is below the industry aver-
age (J. Arrowsmith, personal communication).
Indeed, R&D seems to be especially vulnerable to
the negative impact of mergers and acquisitions. Having
a sense of how mergers occur in R&D organizations
is helpful for understanding this impact. R&D organi-
zations will be the last part of the companies to begin
merger discussions before regulatory approval because
of the commercial sensitivity of the pipeline and the
intellectual property of the company. And when the
discussions about integrating the R&D organiza-
tions finally occur, the initial focus is on Phase III
programmes, followed by mid-stage candidates, with
the early-stage discovery programmes handled last.
These reviews are extensive and time-consuming, as
they require careful consideration of scientific issues
such as efficacy and safety data for each programme,
as well as commercial issues such as potential duplica-
tion and strategic directions of the merged company. In
addition, research organizations often differ proce-
durally in some fundamental processes such as IT
platforms, data handling or adverse event monitoring.
Establishing which system to use or creating a hybrid
takes substantial time for decision-making as well as
implementation.
It is easy to see how early-stage R&D will be slowed
in such situations, as during this period which can
take at least 9months generally no new programmes
are started and hiring will be frozen. Undergoing one
merger will have a substantial negative impact on the
momentum of research programmes, but enduring this
multiple times can be crippling.
Social consequences
For individual employees of companies that are in the
process of merging, uncertainty about their future is
often high. It is hard to quantify the impact that such
uncertainty has on productivity, except that it is nega-
tive, and probably strongly so in many cases. Leaders
of organizations who have completed multiple mergers
may express the view: Weve done this before and we
know how to do it. Mechanically, this may be true, but
for many employees who survive mergers, the thought of
repeating the exercise is not embraced, and could prove
to be numbing to their motivation.
Going forward
It is unlikely that the era of large mergers in the phar-
maceutical industry has ended. However, as a strategy
to achieve top-line growth, it has a clear flaw if the
R&D engine is not growing robustly enough to keep
pace, more mergers and acquisitions will be needed to
continuegrowth.
Not all CEOs of major pharmaceutical companies
believe in this strategy; for example, John Lechleiter,
CEO of Lilly, has stated his opposition to large-scale
combination. And not all CEOs of merged companies
believe in cutting back R&D; for example, Mercks new
CEO, Kenneth Frazier, has recently stated that Merck
will focus on investing in drug development to drive
growth. Whether other leaders in the pharmaceutical
industry will come around to the views of Lechleiter
and Frazier remains to be seen. However, the experience
from the past decade on the negative impact of merg-
ers and acquisitions on R&D productivity should make
these leaders pause when considering major mergers in
thefuture.
Such mergers should also be concerning to patients,
physicians and payers, particularly bearing in mind
recent cutbacks in areas of research such as antibacte-
rial drugs and neuroscience. Industry consolidation has
resulted in less competition and less investment in R&D.
At a time when there is a major need for new treatments
for conditions such as Alzheimers disease, drug-resistant
infections and diabetes, such a trend is alarming.
1. Paul, S. M. et al. How to improve R&D productivity: the
pharmaceutical industrys grand challenge. Nature Rev. Drug
Discov. 9, 203214 (2010).
2. Garnier, J.-P. Rebuilding the R&D engine in Big Pharma. Harvard
Bus. Rev. 86, 6876 (2008).
3. Munos, B. Lessons from 60years of pharmaceutical innovation.
The author declares no competing financial interests.
560 | AUGUST 2011 | VOLUME 10 www.nature.com/reviews/drugdisc
COMMENT
nrd_3514_aug11.indd 560 19/07/2011 15:03
First published in Nature Reviews Drug Discovery 10, 241-242 (April 2011); doi:10.1038/nrd3412

A gUi DE To DRUg Di SCovERy opi ni on
The future of drug development:
advancing clinical trial design
John Orloff, Frank Douglas, Jose Pinheiro, Susan Levinson, Michael Branson,
Pravin Chaturvedi, Ene Ette, Paul Gallo, Gigi Hirsch, Cyrus Mehta, Nitin Patel,
Sameer Sabir, Stacy Springs, Donald Stanski, Matthias R. Evers, Edd Fleming,
Navjot Singh, Tony Tramontin and Howard Golub
Abstract | Declining pharmaceutical industry productivity is well recognized by drug
developers, regulatory authorities and patient groups. A key part of the problem is
that clinical studies are increasingly expensive, driven by the rising costs of
conducting Phase ii and iii trials. it is therefore crucial to ensure that these phases
of drug development are conducted more efficiently and cost-effectively, and that
attrition rates are reduced. in this article, we argue that moving from the traditional
clinical development approach based on sequential, distinct phases towards a
more integrated view that uses adaptive design tools to increase flexibility and
maximize the use of accumulated knowledge could have an important role in
achieving these goals. Applications and examples of the use of these tools such
as Bayesian methodologies in early- and late-stage drug development are
discussed, as well as the advantages, challenges and barriers to their more
widespread implementation.
Pharmaceutical innovation is increasingly
risky, costly and at times inefficient, which
has led to decreased industry productivity
13
.
estimates for the average cost of bringing
a new drug to market range between $800
million and $2 billion, in which late-stage
failures and the rising costs of Phase II
and III trials represent key components
49
.
Conducting these phases of development
more effectively and reducing attrition rates
are therefore major goals. The problem of
attrition is particularly acute in Phase II
trials
10
, owing to factors such as the lack of
proof of relevance for the biological target in
a given disease intervention and insufficient
understanding of the doseresponse
relationship of the new molecular entity.
as recognized by the US Food and drug
administration (Fda) Critical Path Initiative,
novel approaches to clinical trial and
programme design could have a key role in
overcoming these challenges. The traditional
approach to drug development separates
clinical development into sequential,
distinct phases, in which progress is
measured at discrete milestones, separated
by white space. We argue that the effective-
ness of the clinical development can be
improved by adopting a more integrated
model that increases flexibility and maxi-
mizes the use of accumulated knowledge.
In this model, broader, more flexible phases
leading to submission for approval are
designated exploratory and confirmatory
(FIG. 1). This model is adaptive, parallel
and data-led, and allows all available
knowledge to be appropriately shared
across the breadth of development studies
to improve the quality, timeliness and
efficiency of the process.
Central to this model of drug develop-
ment are novel tools, including modelling
and simulation, Bayesian methodologies, and
adaptive designs, such as seamless adaptive
designs and sample-size re-estimation
methods (BOX 1). These can ensure the judi-
cious use of limited patient resources, reduce
patient exposure to ineffective or poorly
tolerated doses, and lead to the recruitment
of patients who, on the basis of biomarker
analysis, are most likely to respond and
those with the most favourable benefit/
risk ratio.
In this article, we describe the general
issues and methods involved, and illustrate
how the tools can be applied in both explora-
tory and confirmatory drug development by
using specific cases in which modern trial
designs and statistical approaches have been
successful. We hope to raise awareness of
these issues among those involved in clinical
trials and provide guidelines to ensure that
the most appropriate solutions are imple-
mented, with the ultimate goal of increasing
the efficiency and probability of success in
clinical development.
Exploratory phase of development
Modelling is a key feature of the more
integrated approach to drug development
(FIG. 1). Biological modelling is used to under-
stand genetic, biochemical and physiological
networks, as well as pathways and processes
underlying disease and pharmacotherapy
11,12
.
Pharmacological modelling guides clinical
trial design, dose selection and development
strategies
13,14
. Finally, statistical modelling can
be used to assess development strategies and
trial designs in populations
11,12,15
. These three
types of modelling should be used throughout
the drug development process to maximize
their impact and synergies.
In the exploratory phase, modelling and
simulation can help refine dose selection
and study design. early development studies
are conducted with fairly restricted resources
(duration, sample sizes and so on), and the
use of all available information is crucial
for effective decision making
16
. However,
it should be noted that early development
decisions based on biomarkers that have
not been fully qualified can be misguided if
such biomarkers eventually do not prove to
correlate with, or be predictive of, the final
outcome. accordingly, it is important to
conduct methodology research in parallel
to the development programme to establish
the correlation between the biomarker and
late-stage endpoints or outcomes.
Modelling and simulation approaches
can be used to represent doseresponse
and timeresponse behaviour of safety and
efficacy endpoints. Furthermore, these
approaches can be combined with Bayesian
methods to provide a continuous flow
of information across different phases of
development. For example, preclinical
data can be used to construct models and
to provide prior information on model
parameters. likewise, the results from a
proof-of-concept (PoC) study can be used to
form prior distributions for a similar model
to be used in a subsequent dose-finding
study
11,12,17,18
.
an additional benefit of modelling in
early development is that it allows the use of
external information (for example, baseline
values for safety endpoints) to estimate char-
acteristics of interest about the population.
Given the vast quantity of data from other
development programmes that are available
in most pharmaceutical companies, as well
as current discussions within the industry
PersPecti ves
naTUre revIeWS | Drug Discovery volUMe 8 | deCeMBer 2009 | 949
nrd_persp_dec09.indd 949 16/11/09 13:57:40
by a full proposal that is reviewed in conjunction with
an external panel of experts. Key criteria include target
novelty, medical need, disease association, reagent avail-
ability and structural data. So far, out of 112 proposals
received, 52 projects have passed the triage and out of the
17 heard at full application stage, 13 have been accepted
and collaborative projects have either started or are under
negotiation. Projects are progressed as true collaborations
between MRCT and the academic group and, impor-
tantly, no transfer of IP rights is required any IP that
is developed as part of the collaboration is jointly owned
and any revenue that is generated is split between the two
parties under the terms of a pre-negotiated agreement.
Aiming for novel compounds rather than targets, Eli
Lilly introduced its Phenotypic Drug Discovery pro-
gramme (PD
2
; www.pd2.lilly.com) in June 2009 as a
platform to search for novel chemical compounds with
high structural diversity. Structures can be submitted
via the internet and are evaluated by specific cheminfor-
matics tools to determine whether they meet Eli Lilly-
defined minimum criteria, such as structural novelty or
drug-like properties. If a compound passes the first-line
evaluation, Eli Lilly analyses the compound in a panel of
phenotypic screens to evaluate whether the compound
has activities linked to its key indications (Alzheimers
disease, cancer, diabetes or bone formation). If the screen
provides promising data (which are provided freely to
the investigators), the compound can be evaluated fur-
ther through either in-licensing or collaboration.
Another approach has been taken by GlaxoSmithKline
with their Pharma in Partnership programme (www.
pharmainpartnership.gsk.com), in which the company
posts information on assets that are either potentially
suitable for repurposing because they show clear phar-
macology but have perhaps failed in their initial indi-
cations, or that are novel targets for which additional
information is required to validate the target or identify
the optimal clinical indication. Academics are invited
to register for updates as new molecules or mechanisms
are posted, and to submit proposals to address the key
question posed for that asset. The potential outcomes for
successful proposals are flexible, ranging from a material
transfer agreement for an in vitro study to full adoption
as a GlaxoSmithKline clinical programme. One particu-
lar expectation is that the successful proposals will lead
to research collaborations with a substantial contribu-
tion from the proposer. The programme was launched in
April 2010, and since then five molecules/mechanisms
have been posted. The first successful proposal has been
identified and a collaboration is under negotiation.
Factors for success
To use crowd sourcing successfully, it is critical that the
questions or challenges to be addressed are suitable,
precisely defined and clearly presented, and that what
is expected from potential solvers and offered by the
searching organization is clearly communicated. In the
examples mentioned, the organizations are seeking novel
targets or compounds in clearly defined indications, or
additional therapeutic options for specific, well-defined
compounds. Use of specific questions within the submis-
sion template can guide the proposer to ensure that they
are addressing the key questions. A precise definition
of search terms (for example, what is a target?) avoids
futile efforts. This also means that organizations have
to be sufficiently transparent in explaining what they
are interested in, and they have to overcome concerns
regarding confidentiality of the information provided.
Operational implementation is another key success
factor. An accessible internet submission tool that pro-
vides a simple and transparent submission process is
needed. A good example is the flow chart of the PD
2

initiative. The transparency and speed of the evaluation
procedure are important. In the case of the G4T pro-
gramme, applicants receive feedback on the success of
their application within 8weeks of submission, and for
Pharma in Partnership, feedback is sent within 28days
of the call closing. Low bureaucratic hurdles for both
partners are also needed to make submissions attractive
and manageable, and the IP policy should also be clearly
defined. Pre-negotiated master contracts can facilitate
rapid execution of agreements.
To generate awareness of the call, the initiative has to
be communicated appropriately; for example, through
scientific conferences, targeted advertising and mail-
ings to relevant institutions, societies and field leaders.
Finally, it is essential that proposer organizations are
willing and ready to take up externally generated ideas.
Collaboration with external parties needs particu-
lar attention to relationship management and usually
requires dedicated resources.
Crowd sourcing initiatives in drug discovery are still
in their infancy, and whether they will have a substan-
tial impact has yet to be proven. Nevertheless, an initial
evaluation of the impact of the InnoCentive initiative
was promising
4
, and our own experience so far suggests
that such initiatives could provide an important source
of future innovation in early-stage drug discovery.
Monika Lessl, Justin S.Bryans, Duncan Richards
and Khusru Asadullah
1. Paul, S. M. et al. How to improve R&D productivity: the pharmaceutical
industrys grand challenge. Nature Rev. Drug. Discov. 9, 203214 (2010).
2. Howe, J. (ed.) Crowdsourcing: Why the Power of the Crowd is Driving
the Future of Business. (Crown Publishing Group, New York, 2008).
3. Lessl, M. et al. Grants4Targets a novel approach to translate basic
research to novel drugs. Drug Discov. Today 1Dec 2010
(doi:10.1016/j.drudis.2010.11.013).
4. Bishop, M. The total economic impact of InnoCentives enterprise
solution: challenges, InnoCentive@work and ONRAMP. (Forrester
Business Consulting, Cambridge, Massachusetts, 2010).
The authors declare competing financial interests: see Web version for
details.
Acknowledgements
The authors thank S. Schoepe for her valuable contributions in editing the
manuscript.
242 | APRIL 2011 | VOLUME 10 www.nature.com/reviews/drugdisc
COMMENT
nrd_3412_apr11.indd 242 17/03/2011 11:40

A gUi DE To DRUg Di SCovERy opi ni on
The future of drug development:
advancing clinical trial design
John Orloff, Frank Douglas, Jose Pinheiro, Susan Levinson, Michael Branson,
Pravin Chaturvedi, Ene Ette, Paul Gallo, Gigi Hirsch, Cyrus Mehta, Nitin Patel,
Sameer Sabir, Stacy Springs, Donald Stanski, Matthias R. Evers, Edd Fleming,
Navjot Singh, Tony Tramontin and Howard Golub
Abstract | Declining pharmaceutical industry productivity is well recognized by drug
developers, regulatory authorities and patient groups. A key part of the problem is
that clinical studies are increasingly expensive, driven by the rising costs of
conducting Phase ii and iii trials. it is therefore crucial to ensure that these phases
of drug development are conducted more efficiently and cost-effectively, and that
attrition rates are reduced. in this article, we argue that moving from the traditional
clinical development approach based on sequential, distinct phases towards a
more integrated view that uses adaptive design tools to increase flexibility and
maximize the use of accumulated knowledge could have an important role in
achieving these goals. Applications and examples of the use of these tools such
as Bayesian methodologies in early- and late-stage drug development are
discussed, as well as the advantages, challenges and barriers to their more
widespread implementation.
Pharmaceutical innovation is increasingly
risky, costly and at times inefficient, which
has led to decreased industry productivity
13
.
estimates for the average cost of bringing
a new drug to market range between $800
million and $2 billion, in which late-stage
failures and the rising costs of Phase II
and III trials represent key components
49
.
Conducting these phases of development
more effectively and reducing attrition rates
are therefore major goals. The problem of
attrition is particularly acute in Phase II
trials
10
, owing to factors such as the lack of
proof of relevance for the biological target in
a given disease intervention and insufficient
understanding of the doseresponse
relationship of the new molecular entity.
as recognized by the US Food and drug
administration (Fda) Critical Path Initiative,
novel approaches to clinical trial and
programme design could have a key role in
overcoming these challenges. The traditional
approach to drug development separates
clinical development into sequential,
distinct phases, in which progress is
measured at discrete milestones, separated
by white space. We argue that the effective-
ness of the clinical development can be
improved by adopting a more integrated
model that increases flexibility and maxi-
mizes the use of accumulated knowledge.
In this model, broader, more flexible phases
leading to submission for approval are
designated exploratory and confirmatory
(FIG. 1). This model is adaptive, parallel
and data-led, and allows all available
knowledge to be appropriately shared
across the breadth of development studies
to improve the quality, timeliness and
efficiency of the process.
Central to this model of drug develop-
ment are novel tools, including modelling
and simulation, Bayesian methodologies, and
adaptive designs, such as seamless adaptive
designs and sample-size re-estimation
methods (BOX 1). These can ensure the judi-
cious use of limited patient resources, reduce
patient exposure to ineffective or poorly
tolerated doses, and lead to the recruitment
of patients who, on the basis of biomarker
analysis, are most likely to respond and
those with the most favourable benefit/
risk ratio.
In this article, we describe the general
issues and methods involved, and illustrate
how the tools can be applied in both explora-
tory and confirmatory drug development by
using specific cases in which modern trial
designs and statistical approaches have been
successful. We hope to raise awareness of
these issues among those involved in clinical
trials and provide guidelines to ensure that
the most appropriate solutions are imple-
mented, with the ultimate goal of increasing
the efficiency and probability of success in
clinical development.
Exploratory phase of development
Modelling is a key feature of the more
integrated approach to drug development
(FIG. 1). Biological modelling is used to under-
stand genetic, biochemical and physiological
networks, as well as pathways and processes
underlying disease and pharmacotherapy
11,12
.
Pharmacological modelling guides clinical
trial design, dose selection and development
strategies
13,14
. Finally, statistical modelling can
be used to assess development strategies and
trial designs in populations
11,12,15
. These three
types of modelling should be used throughout
the drug development process to maximize
their impact and synergies.
In the exploratory phase, modelling and
simulation can help refine dose selection
and study design. early development studies
are conducted with fairly restricted resources
(duration, sample sizes and so on), and the
use of all available information is crucial
for effective decision making
16
. However,
it should be noted that early development
decisions based on biomarkers that have
not been fully qualified can be misguided if
such biomarkers eventually do not prove to
correlate with, or be predictive of, the final
outcome. accordingly, it is important to
conduct methodology research in parallel
to the development programme to establish
the correlation between the biomarker and
late-stage endpoints or outcomes.
Modelling and simulation approaches
can be used to represent doseresponse
and timeresponse behaviour of safety and
efficacy endpoints. Furthermore, these
approaches can be combined with Bayesian
methods to provide a continuous flow
of information across different phases of
development. For example, preclinical
data can be used to construct models and
to provide prior information on model
parameters. likewise, the results from a
proof-of-concept (PoC) study can be used to
form prior distributions for a similar model
to be used in a subsequent dose-finding
study
11,12,17,18
.
an additional benefit of modelling in
early development is that it allows the use of
external information (for example, baseline
values for safety endpoints) to estimate char-
acteristics of interest about the population.
Given the vast quantity of data from other
development programmes that are available
in most pharmaceutical companies, as well
as current discussions within the industry
PersPecti ves
by a full proposal that is reviewed in conjunction with
an external panel of experts. Key criteria include target
novelty, medical need, disease association, reagent avail-
ability and structural data. So far, out of 112 proposals
received, 52 projects have passed the triage and out of the
17 heard at full application stage, 13 have been accepted
and collaborative projects have either started or are under
negotiation. Projects are progressed as true collaborations
between MRCT and the academic group and, impor-
tantly, no transfer of IP rights is required any IP that
is developed as part of the collaboration is jointly owned
and any revenue that is generated is split between the two
parties under the terms of a pre-negotiated agreement.
Aiming for novel compounds rather than targets, Eli
Lilly introduced its Phenotypic Drug Discovery pro-
gramme (PD
2
; www.pd2.lilly.com) in June 2009 as a
platform to search for novel chemical compounds with
high structural diversity. Structures can be submitted
via the internet and are evaluated by specific cheminfor-
matics tools to determine whether they meet Eli Lilly-
defined minimum criteria, such as structural novelty or
drug-like properties. If a compound passes the first-line
evaluation, Eli Lilly analyses the compound in a panel of
phenotypic screens to evaluate whether the compound
has activities linked to its key indications (Alzheimers
disease, cancer, diabetes or bone formation). If the screen
provides promising data (which are provided freely to
the investigators), the compound can be evaluated fur-
ther through either in-licensing or collaboration.
Another approach has been taken by GlaxoSmithKline
with their Pharma in Partnership programme (www.
pharmainpartnership.gsk.com), in which the company
posts information on assets that are either potentially
suitable for repurposing because they show clear phar-
macology but have perhaps failed in their initial indi-
cations, or that are novel targets for which additional
information is required to validate the target or identify
the optimal clinical indication. Academics are invited
to register for updates as new molecules or mechanisms
are posted, and to submit proposals to address the key
question posed for that asset. The potential outcomes for
successful proposals are flexible, ranging from a material
transfer agreement for an in vitro study to full adoption
as a GlaxoSmithKline clinical programme. One particu-
lar expectation is that the successful proposals will lead
to research collaborations with a substantial contribu-
tion from the proposer. The programme was launched in
April 2010, and since then five molecules/mechanisms
have been posted. The first successful proposal has been
identified and a collaboration is under negotiation.
Factors for success
To use crowd sourcing successfully, it is critical that the
questions or challenges to be addressed are suitable,
precisely defined and clearly presented, and that what
is expected from potential solvers and offered by the
searching organization is clearly communicated. In the
examples mentioned, the organizations are seeking novel
targets or compounds in clearly defined indications, or
additional therapeutic options for specific, well-defined
compounds. Use of specific questions within the submis-
sion template can guide the proposer to ensure that they
are addressing the key questions. A precise definition
of search terms (for example, what is a target?) avoids
futile efforts. This also means that organizations have
to be sufficiently transparent in explaining what they
are interested in, and they have to overcome concerns
regarding confidentiality of the information provided.
Operational implementation is another key success
factor. An accessible internet submission tool that pro-
vides a simple and transparent submission process is
needed. A good example is the flow chart of the PD
2

initiative. The transparency and speed of the evaluation
procedure are important. In the case of the G4T pro-
gramme, applicants receive feedback on the success of
their application within 8weeks of submission, and for
Pharma in Partnership, feedback is sent within 28days
of the call closing. Low bureaucratic hurdles for both
partners are also needed to make submissions attractive
and manageable, and the IP policy should also be clearly
defined. Pre-negotiated master contracts can facilitate
rapid execution of agreements.
To generate awareness of the call, the initiative has to
be communicated appropriately; for example, through
scientific conferences, targeted advertising and mail-
ings to relevant institutions, societies and field leaders.
Finally, it is essential that proposer organizations are
willing and ready to take up externally generated ideas.
Collaboration with external parties needs particu-
lar attention to relationship management and usually
requires dedicated resources.
Crowd sourcing initiatives in drug discovery are still
in their infancy, and whether they will have a substan-
tial impact has yet to be proven. Nevertheless, an initial
evaluation of the impact of the InnoCentive initiative
was promising
4
, and our own experience so far suggests
that such initiatives could provide an important source
of future innovation in early-stage drug discovery.
Monika Lessl, Justin S.Bryans, Duncan Richards
and Khusru Asadullah
1. Paul, S. M. et al. How to improve R&D productivity: the pharmaceutical
industrys grand challenge. Nature Rev. Drug. Discov. 9, 203214 (2010).
2. Howe, J. (ed.) Crowdsourcing: Why the Power of the Crowd is Driving
the Future of Business. (Crown Publishing Group, New York, 2008).
3. Lessl, M. et al. Grants4Targets a novel approach to translate basic
research to novel drugs. Drug Discov. Today 1Dec 2010
(doi:10.1016/j.drudis.2010.11.013).
4. Bishop, M. The total economic impact of InnoCentives enterprise
solution: challenges, InnoCentive@work and ONRAMP. (Forrester
Business Consulting, Cambridge, Massachusetts, 2010).
The authors declare competing financial interests: see Web version for
details.
Acknowledgements
The authors thank S. Schoepe for her valuable contributions in editing the
manuscript.
242 | APRIL 2011 | VOLUME 10 www.nature.com/reviews/drugdisc
COMMENT
nrd_3412_apr11.indd 242 17/03/2011 11:40
First published in Nature Reviews Drug Discovery 8, 949-957 (December 2009); doi:10.1038/nrd3025
The standard approach to early develop-
ment programmes is to separate the trials
for PoC, dose ranging and dose selection.
adaptive designs offer several benefits over
the standard approach. For example, a PoC
trial can be combined with a dose-ranging
trial (BOX 2). This approach has distinct
advantages, in that it reduces start-up costs
and the time between trials, and potentially
increases statistical power and improves
estimates of doseresponse. adaptive
designs can also enable trialists to work
with more candidate doses without increasing
sample size. This is important to reduce
risk of failure in confirmatory trials, where
it has been estimated that, industry-wide,
45% of Phase III programmes do not have
the optimum dose
3
. adaptive dose-ranging
studies are discussed further in BOX 3
(REFS 27,28).
There are a number of requirements
for successful implementation of adaptive
trial designs
2326
. drug responses need to be
rapidly observable relative to accrual rate;
alternatively, good longitudinal models
can be used to forecast endpoints in time
to adapt dose assignments for future sub-
jects (assuming, of course, that the early
measurements are good predictors of the
late endpoint values). adaptive trials also
necessitate more up-front statistical work
to model doseresponse curves and to
perform simulations and many
simulations are required to find the best
combinations of sample size, the randomi-
zation ratio between placebo and drug,
starting dose and number of doses.
This in turn demands efficient program-
ming to develop complex algorithms and
fast computing platforms.
Confirmatory phase of development
The primary goals of a confirmatory clinical
trial are to ensure that the diagnostic or
therapeutic intervention causes less harm
than good (safety) and to efficiently and
confidently find the actual effect size on
the chosen primary outcome(s) within the
identified patient population (efficacy).
optimization of trial design during con-
firmatory development holds the promise
of greater success rates, improved efficiency,
better detection of safety signals, compressed
timelines, smaller overall programmes and
lower attrition rates. a number of novel
approaches to confirmatory development
that can contribute to fulfilling this promise
are highlighted below.
Seamless adaptive designs. efficiency of the
drug development process can be increased
through the use of seamless adaptive
designs, which aim to combine objectives
traditionally addressed in separate trials
into a single trial
25,29
. a specific example is
the seamless adaptive Phase II/III design
addressing objectives normally achieved
through separate Phase II and III trials.
These trials are confirmatory in nature,
as opposed to seamless adaptive trials in
early development, which are essentially
exploratory. The first stage of a seamless
adaptive Phase II/III trial might be similar
to a late-Phase II trial, with a control group
and several treatment groups (for example,
different dose levels of the same treatment).
results are examined at the end of the first
stage, and one or more of the treatment
groups are selected to continue, along with
the control group, into the trials second
stage. The final analysis comparing the

Box 1 | Tools, methods and designs for enhancing clinical development
Here, we summarize some of the key tools, methods and designs that can be incorporated into the
drug development process.
Modelling and simulation
These techniques are a cornerstone of the novel drug development model. In the exploratory
phase, modelling and simulation can help refine dose selection and study design, and to represent
doseresponse and timeresponse behaviour of safety and efficacy endpoints. In combination
with Bayesian methods, these can provide a continuous flow of information across different
phases of development. Modelling in early development also enables the use of external
information (an important issue in light of current discussions within the industry about sharing
placebo data across companies), which could greatly increase the efficiency of investigations in
early development.
In the confirmatory phase, simulation can clarify how different study designs affect the outcome
and likelihood of success, thereby guiding development strategy. In the latter case, this is
facilitated by pooling many sources of data both from prior studies of the drug and external data
that might be an informative guide to achieve better decision-making. Furthermore, these
techniques can be used not just during the trial-design process, but also mid-study through the
use of adaptive trial designs.
Bayesian methodology
This relies on the use of probability models to describe knowledge about parameters of interest
(for example, the treatment effect of a drug in development). Bayesian inference uses principles
from the scientific method to combine prior beliefs with observed data, producing enhanced,
updated information (for reviews, see REFS 22,23). Using Bayesian methodologies, initial beliefs
about the parameters are summarized in their prior distribution. Then, new data values are
collected experimentally (for example, patient survival in an oncology trial) and the probability
distribution of these values leads to the likelihood function (the observed evidence on the
parameters). The two elements are then combined, using Bayes theorem, to produce the posterior
distribution of the parameters that is, the updated knowledge given the observed evidence.
By contrast, frequentist methods rely solely on observed evidence for inferences, and typically do
not formally take into account prior information.
Adaptive designs
In adaptive trial designs, interim data from a trial is used to modify and improve the study design,
in a pre-planned manner and without undermining its validity or integrity. In the exploratory
setting, an adaptive trial can assign a larger proportion of the enrolled subjects to the treatment
arms that are performing well, drop arms that are performing poorly, and investigate a wider range
of doses so as to more effectively select doses that are most likely to succeed in the confirmatory
phase. In the confirmatory phase, adaptive design can facilitate the early identification of
efficacious treatments, decisions to drop poorly performing trial arms, determining whether the
trial should be terminated for futility and making sample-size adjustments at interim time points
to ensure that the trial is adequately powered. In some cases, it might even be possible to enrich
the patient population by altering the eligibility criteria at an interim time point.
seamless designs
Such designs combine, in a single trial, the objectives that are traditionally addressed in separate
trials. A seamless adaptive design addresses objectives normally achieved through separate trials
using data from all trial stages, such as seamless adaptive Phase II/III trials.
sample size re-estimation methods
These provide the flexibility to either increase or decrease the sample size at an interim point in
the trial. This is important in cases in which there is uncertainty about between-subject variance
in the response or uncertainty about the clinically meaningful effect size at which to power the
trial. These methods allow the study to begin with a certain sample size that can be increased or
decreased at an interim point, and even allow for an efficacy-stopping boundary.
PersPecti ves
Target PoC Approval
Apply innovative tools and clinical
trial designs such as adaptive or
seamless studies
Identify target patient population,
confirm optimal dose and
dosing regimen and establish the
benefit/risk ratio
Apply biomarkers, modelling and
simulation, and advanced statistical
methodology
Demonstrate PoC and establish
dose selection
Exploratory phase Confirmatory phase
Target discovery
and validation
PoC clinical trials Clinical development
about sharing placebo data across compa-
nies, this has huge potential for improving
the efficiency of investigation in early
development.
Modelling and simulation for dose and
dose regimen selection. an important goal
of a drug development programme is the
selection of a dose and dosing regimen that
achieves the target clinical benefit while
minimizing undesirable adverse effects.
Biological and pharmacological modelling
can be very useful in this context
19,20
. For
example, we (J.o., J.P., M.B., P.G. and d.S.)
have used such modelling in the dose selec-
tion for canakinumab (Ilaris; novartis),
a monoclonal antibody that has recently
been approved for the treatment of the rare
genetic disease MuckleWells syndrome
(FIG. 2). Clinical data on the relationship
between activity of the therapeutic target
(interleukin-1), markers of inflammation
and remission of symptoms were captured
in a mathematical model that was con-
tinuously adjusted to fit emerging data.
Simulation was then used to propose a suit-
able dose and dosing regimen to achieve the
desired response for the majority of patients
in this instance, an 80% probability that
90% of patients would remain flare-free
for 2 months. The data derived from this
modelling exercise allowed for selection of
a dosing regimen that was investigated and
confirmed in a Phase III trial
21
(clinical data
on various dosing intervals provided the
raw data for the modelling and simulation
exercise that finalized the dose and regimen
selection for Phase III). Similarly, model-
ling has been used to predict the impact
of changing the dose or dosing regimen of
a dipeptidyl peptidase Iv inhibitor that
is being developed for the treatment
of type 2 diabetes (see Supplementary
information S1 (box)).
Bayesian modelling combined with use of
external baseline data to improve efficacy and
safety signal detection in early development.
early development studies for establish-
ing PoC often use small patient cohorts
(1020 subjects). These patients are typically
observed for a relatively short period of time
(several weeks) to evaluate early efficacy and
safety signals, which are frequently meas-
ured on a continuous scale and observed
several times over the duration of the study.
However, the endpoints for the decision to
proceed with development or not are typically
based on a single time point (for example,
change from baseline at the end of the study)
and use dichotomized versions of the original
variables to characterize responder and
non-responder behaviour. an example of
the latter is the transformation of continuous
liver function test measurements (for example,
alanine aminotransferase (alT) and
aspartate aminotransferase (aST)) into
binary indicators (for instance, exceeding
three times the upper limit of normal
(Uln)). There are, therefore, two types of
information loss that often occur in PoC
studies: the dichotomization of continuous
endpoints and a failure to use all of the
available longitudinal measurements
collected in the study
22
.
a typical design for efficacy and safety
evaluation in a PoC study is to use cohorts
in a dose-escalation algorithm. Cohorts
are assigned, in sequence, to increasing
doses until the maximum tolerated dose is
reached, or unacceptable safety is observed
for a given cohort. a new cohort is only
allowed to start once acceptable safety
signals are verified for all previous doses.
at the end of the study, the goal is to
either determine a dose range for further
exploration in Phase IIb, or to conclude
that no PoC can be established based on
the efficacysafety trade-off.
Because of small cohort sizes, only safety
problems occurring in a relatively large per-
centage of patients can be reliably detected
by dose-escalation procedures. likewise,
only relatively strong efficacy signals can be
detected with reasonable statistical power.
The detection of safety and efficacy signals
can be made more efficient in various ways:
by drawing on data and information external
to the trial, and by deploying longitudinal
modelling approaches to make use of all
available information. Furthermore, the utility
of PoC studies within drug development
programmes can be enhanced by incorpo-
rating the information obtained in them
directly into later-phase trials
11,12
. Bayesian
modelling techniques are particularly useful
in implementing these approaches. a PoC
study in dyslipidaemia that illustrates the
methods mentioned above is provided in
Supplementary information S2 (box)).
Adaptive trial designs in early development.
The core concept of adaptive trial design
(also known as flexible design) is that it uses
accumulating data to decide on how to
modify aspects of the study mid-trial, in a
pre-planned manner, without undermining
the validity or integrity of the study
2326
.
Possible adaptations include adjustments
to sample size, allocation of treatments,
the addition or deletion of treatment arms,
inclusion and exclusion criteria for the study
population, adjusting statistical hypotheses
(such as non-inferiority or superiority),
and combining trials or treatment phases.
adaptive trials have the potential to translate
into more ethical treatment of patients within
trials, more efficient drug development and
better use of available resources.
Figure 1 | A novel model for clinical development. During the exploratory phase of development,
this model uses all available knowledge and tools, including biomarkers, modelling and simulation, as
well as advanced statistical methodology. Trials are designed to determine proof-of-concept (Poc)
and to establish dose selection to a level of rigour that will enhance the likelihood of success in the
confirmatory phase. During the confirmatory phase, modern designs, tools and knowledge are applied
to larger-scale studies with the goal of identifying the target patient population in which the drug is
efficacious, establishing the benefit/risk ratio and confirming the optimal dose and dosing regimen.
During this phase, innovative clinical trial designs such as adaptive or seamless studies compress time-
lines, improve dose and regimen selection, and reduce the number of patients assigned to non-viable
dosing regimens.
PersPecti ves
950 | deCeMBer 2009 | volUMe 8 www.nature.com/reviews/drugdisc
The standard approach to early develop-
ment programmes is to separate the trials
for PoC, dose ranging and dose selection.
adaptive designs offer several benefits over
the standard approach. For example, a PoC
trial can be combined with a dose-ranging
trial (BOX 2). This approach has distinct
advantages, in that it reduces start-up costs
and the time between trials, and potentially
increases statistical power and improves
estimates of doseresponse. adaptive
designs can also enable trialists to work
with more candidate doses without increasing
sample size. This is important to reduce
risk of failure in confirmatory trials, where
it has been estimated that, industry-wide,
45% of Phase III programmes do not have
the optimum dose
3
. adaptive dose-ranging
studies are discussed further in BOX 3
(REFS 27,28).
There are a number of requirements
for successful implementation of adaptive
trial designs
2326
. drug responses need to be
rapidly observable relative to accrual rate;
alternatively, good longitudinal models
can be used to forecast endpoints in time
to adapt dose assignments for future sub-
jects (assuming, of course, that the early
measurements are good predictors of the
late endpoint values). adaptive trials also
necessitate more up-front statistical work
to model doseresponse curves and to
perform simulations and many
simulations are required to find the best
combinations of sample size, the randomi-
zation ratio between placebo and drug,
starting dose and number of doses.
This in turn demands efficient program-
ming to develop complex algorithms and
fast computing platforms.
Confirmatory phase of development
The primary goals of a confirmatory clinical
trial are to ensure that the diagnostic or
therapeutic intervention causes less harm
than good (safety) and to efficiently and
confidently find the actual effect size on
the chosen primary outcome(s) within the
identified patient population (efficacy).
optimization of trial design during con-
firmatory development holds the promise
of greater success rates, improved efficiency,
better detection of safety signals, compressed
timelines, smaller overall programmes and
lower attrition rates. a number of novel
approaches to confirmatory development
that can contribute to fulfilling this promise
are highlighted below.
Seamless adaptive designs. efficiency of the
drug development process can be increased
through the use of seamless adaptive
designs, which aim to combine objectives
traditionally addressed in separate trials
into a single trial
25,29
. a specific example is
the seamless adaptive Phase II/III design
addressing objectives normally achieved
through separate Phase II and III trials.
These trials are confirmatory in nature,
as opposed to seamless adaptive trials in
early development, which are essentially
exploratory. The first stage of a seamless
adaptive Phase II/III trial might be similar
to a late-Phase II trial, with a control group
and several treatment groups (for example,
different dose levels of the same treatment).
results are examined at the end of the first
stage, and one or more of the treatment
groups are selected to continue, along with
the control group, into the trials second
stage. The final analysis comparing the

Box 1 | Tools, methods and designs for enhancing clinical development
Here, we summarize some of the key tools, methods and designs that can be incorporated into the
drug development process.
Modelling and simulation
These techniques are a cornerstone of the novel drug development model. In the exploratory
phase, modelling and simulation can help refine dose selection and study design, and to represent
doseresponse and timeresponse behaviour of safety and efficacy endpoints. In combination
with Bayesian methods, these can provide a continuous flow of information across different
phases of development. Modelling in early development also enables the use of external
information (an important issue in light of current discussions within the industry about sharing
placebo data across companies), which could greatly increase the efficiency of investigations in
early development.
In the confirmatory phase, simulation can clarify how different study designs affect the outcome
and likelihood of success, thereby guiding development strategy. In the latter case, this is
facilitated by pooling many sources of data both from prior studies of the drug and external data
that might be an informative guide to achieve better decision-making. Furthermore, these
techniques can be used not just during the trial-design process, but also mid-study through the
use of adaptive trial designs.
Bayesian methodology
This relies on the use of probability models to describe knowledge about parameters of interest
(for example, the treatment effect of a drug in development). Bayesian inference uses principles
from the scientific method to combine prior beliefs with observed data, producing enhanced,
updated information (for reviews, see REFS 22,23). Using Bayesian methodologies, initial beliefs
about the parameters are summarized in their prior distribution. Then, new data values are
collected experimentally (for example, patient survival in an oncology trial) and the probability
distribution of these values leads to the likelihood function (the observed evidence on the
parameters). The two elements are then combined, using Bayes theorem, to produce the posterior
distribution of the parameters that is, the updated knowledge given the observed evidence.
By contrast, frequentist methods rely solely on observed evidence for inferences, and typically do
not formally take into account prior information.
Adaptive designs
In adaptive trial designs, interim data from a trial is used to modify and improve the study design,
in a pre-planned manner and without undermining its validity or integrity. In the exploratory
setting, an adaptive trial can assign a larger proportion of the enrolled subjects to the treatment
arms that are performing well, drop arms that are performing poorly, and investigate a wider range
of doses so as to more effectively select doses that are most likely to succeed in the confirmatory
phase. In the confirmatory phase, adaptive design can facilitate the early identification of
efficacious treatments, decisions to drop poorly performing trial arms, determining whether the
trial should be terminated for futility and making sample-size adjustments at interim time points
to ensure that the trial is adequately powered. In some cases, it might even be possible to enrich
the patient population by altering the eligibility criteria at an interim time point.
seamless designs
Such designs combine, in a single trial, the objectives that are traditionally addressed in separate
trials. A seamless adaptive design addresses objectives normally achieved through separate trials
using data from all trial stages, such as seamless adaptive Phase II/III trials.
sample size re-estimation methods
These provide the flexibility to either increase or decrease the sample size at an interim point in
the trial. This is important in cases in which there is uncertainty about between-subject variance
in the response or uncertainty about the clinically meaningful effect size at which to power the
trial. These methods allow the study to begin with a certain sample size that can be increased or
decreased at an interim point, and even allow for an efficacy-stopping boundary.
PersPecti ves
Target PoC Approval
Apply innovative tools and clinical
trial designs such as adaptive or
seamless studies
Identify target patient population,
confirm optimal dose and
dosing regimen and establish the
benefit/risk ratio
Apply biomarkers, modelling and
simulation, and advanced statistical
methodology
Demonstrate PoC and establish
dose selection
Exploratory phase Confirmatory phase
Target discovery
and validation
PoC clinical trials Clinical development
about sharing placebo data across compa-
nies, this has huge potential for improving
the efficiency of investigation in early
development.
Modelling and simulation for dose and
dose regimen selection. an important goal
of a drug development programme is the
selection of a dose and dosing regimen that
achieves the target clinical benefit while
minimizing undesirable adverse effects.
Biological and pharmacological modelling
can be very useful in this context
19,20
. For
example, we (J.o., J.P., M.B., P.G. and d.S.)
have used such modelling in the dose selec-
tion for canakinumab (Ilaris; novartis),
a monoclonal antibody that has recently
been approved for the treatment of the rare
genetic disease MuckleWells syndrome
(FIG. 2). Clinical data on the relationship
between activity of the therapeutic target
(interleukin-1), markers of inflammation
and remission of symptoms were captured
in a mathematical model that was con-
tinuously adjusted to fit emerging data.
Simulation was then used to propose a suit-
able dose and dosing regimen to achieve the
desired response for the majority of patients
in this instance, an 80% probability that
90% of patients would remain flare-free
for 2 months. The data derived from this
modelling exercise allowed for selection of
a dosing regimen that was investigated and
confirmed in a Phase III trial
21
(clinical data
on various dosing intervals provided the
raw data for the modelling and simulation
exercise that finalized the dose and regimen
selection for Phase III). Similarly, model-
ling has been used to predict the impact
of changing the dose or dosing regimen of
a dipeptidyl peptidase Iv inhibitor that
is being developed for the treatment
of type 2 diabetes (see Supplementary
information S1 (box)).
Bayesian modelling combined with use of
external baseline data to improve efficacy and
safety signal detection in early development.
early development studies for establish-
ing PoC often use small patient cohorts
(1020 subjects). These patients are typically
observed for a relatively short period of time
(several weeks) to evaluate early efficacy and
safety signals, which are frequently meas-
ured on a continuous scale and observed
several times over the duration of the study.
However, the endpoints for the decision to
proceed with development or not are typically
based on a single time point (for example,
change from baseline at the end of the study)
and use dichotomized versions of the original
variables to characterize responder and
non-responder behaviour. an example of
the latter is the transformation of continuous
liver function test measurements (for example,
alanine aminotransferase (alT) and
aspartate aminotransferase (aST)) into
binary indicators (for instance, exceeding
three times the upper limit of normal
(Uln)). There are, therefore, two types of
information loss that often occur in PoC
studies: the dichotomization of continuous
endpoints and a failure to use all of the
available longitudinal measurements
collected in the study
22
.
a typical design for efficacy and safety
evaluation in a PoC study is to use cohorts
in a dose-escalation algorithm. Cohorts
are assigned, in sequence, to increasing
doses until the maximum tolerated dose is
reached, or unacceptable safety is observed
for a given cohort. a new cohort is only
allowed to start once acceptable safety
signals are verified for all previous doses.
at the end of the study, the goal is to
either determine a dose range for further
exploration in Phase IIb, or to conclude
that no PoC can be established based on
the efficacysafety trade-off.
Because of small cohort sizes, only safety
problems occurring in a relatively large per-
centage of patients can be reliably detected
by dose-escalation procedures. likewise,
only relatively strong efficacy signals can be
detected with reasonable statistical power.
The detection of safety and efficacy signals
can be made more efficient in various ways:
by drawing on data and information external
to the trial, and by deploying longitudinal
modelling approaches to make use of all
available information. Furthermore, the utility
of PoC studies within drug development
programmes can be enhanced by incorpo-
rating the information obtained in them
directly into later-phase trials
11,12
. Bayesian
modelling techniques are particularly useful
in implementing these approaches. a PoC
study in dyslipidaemia that illustrates the
methods mentioned above is provided in
Supplementary information S2 (box)).
Adaptive trial designs in early development.
The core concept of adaptive trial design
(also known as flexible design) is that it uses
accumulating data to decide on how to
modify aspects of the study mid-trial, in a
pre-planned manner, without undermining
the validity or integrity of the study
2326
.
Possible adaptations include adjustments
to sample size, allocation of treatments,
the addition or deletion of treatment arms,
inclusion and exclusion criteria for the study
population, adjusting statistical hypotheses
(such as non-inferiority or superiority),
and combining trials or treatment phases.
adaptive trials have the potential to translate
into more ethical treatment of patients within
trials, more efficient drug development and
better use of available resources.
Figure 1 | A novel model for clinical development. During the exploratory phase of development,
this model uses all available knowledge and tools, including biomarkers, modelling and simulation, as
well as advanced statistical methodology. Trials are designed to determine proof-of-concept (Poc)
and to establish dose selection to a level of rigour that will enhance the likelihood of success in the
confirmatory phase. During the confirmatory phase, modern designs, tools and knowledge are applied
to larger-scale studies with the goal of identifying the target patient population in which the drug is
efficacious, establishing the benefit/risk ratio and confirming the optimal dose and dosing regimen.
During this phase, innovative clinical trial designs such as adaptive or seamless studies compress time-
lines, improve dose and regimen selection, and reduce the number of patients assigned to non-viable
dosing regimens.
PersPecti ves
M
e
a
n

r
e
s
p
o
n
s
e
25
20
15
10
5
0
6 8 7 5 4 3 2
Dose
Scenario 1
Scenario 2
Scenario 3
1
a
Scenario 4
Scenario 5
Scenario 6
Scenario 7
A
v
e
r
a
g
e

a
l
l
o
c
a
t
i
o
n
20
18
16
12
14
10
8
6
4
2
0
Dose
Average allocation
Efficiency
c
b
E
f
f
i
c
i
e
n
c
y
E
f
f
i
c
i
e
n
c
y
10
9
8
6
7
5
4
3
2
1
0
10
9
8
6
7
5
4
3
2
1
0
6 7 5 4 3 2
Dose
Adaptive design (sample size = 120)
Standard design (sample size = 120)
1
d
1 2 3 4 5 6 7
Scenario Power of
standard
PoC trial*
Power of
combined
adaptive design
1
2
3
4
5
6
7
5
33
32
86
85
100
99
5
84
67
100
99
100
100
*Sample size = 30.
versus the selected treatment that do not
account for the design will not be appropriate.
Finally, the appropriateness of the design
does not depend on any particular algorithm
for choosing the patient group to be contin-
ued; it is not even necessary for a firm
algorithm to be specified in advance, although
the general principles that will govern the
decision should be clear in advance.
Sample size re-estimation within a
confirmatory trial (Phase III). Sample size
re-estimation (SSr) provides a mechanism for
appropriately using the information obtained
during a confirmatory study to inform and
adjust the necessary sample size going for-
ward
30,31
. This process increases confidence
that an appropriate sample size has been
chosen to answer the primary study questions.
The standard approach used to power a
confirmatory study is to first estimate the
underlying treatment effect on the primary
endpoint based on available prior informa-
tion. The parameter denotes the true
underlying difference between the treat-
ment and control arms with respect to the
primary endpoint. even though the true
value of is unknown, the trial investigators
will usually have in mind a specific value,
min
, which represents the smallest clinically
important delta (SCId) for this clinical trial.
next, the trial designers will determine the
sample size that can detect values of , based
on prior information, that exceed the SCId
with good power. The standard deviation
(between subject variability) is a nuisance
parameter whose true value must be
estimated in order to proceed with the
sample size calculation.
The SCId can often be pre-specified
from purely clinical arguments, whereas the
actual effect size is unknown. Therefore, it is
possible in principle to design a study with
a fixed sample size that will have adequate
power to detect the SCId, in the absence
of adequate prior information about the
actual effect size of the test agent. This is
what statisticians envisaged when they
created the fixed-sample methodology.
However, this fixed sample methodology
has several drawbacks. If the actual effect is
substantially larger than the SCId, a smaller
sample size would have sufficed to attain
adequate power
32
.
Sponsors will not often risk significant
resources on trial sizes based on SCId
assumptions that would lead to larger trials
than the current best guess about the actual
effect size (BOX 4). Instead, a smaller trial
corresponding to that best guess may be run;
if that assumption is too optimistic, and the
Box 2 | Case study: combining poC and dose-ranging trials into a single adaptive trial
This example illustrates how a single adaptive trial can replace two standard trials
proof-of-concept (PoC) and dose-ranging and that the combined trial has greater power than
the standard PoC design, and is substantially better at estimating the doseresponse curve.
The trial evaluated an analgesic drug to treat dental pain and tested seven doses of the drug.
Several designs with different sample sizes, randomization ratios of drug to placebo and starting
doses were simulated against several scenarios. Here, we describe one design with a sample
size of 120 subjects (40 placebo, 80 drug). Bayesian adaptive trials were simulated over seven
drugresponse scenarios to enable comparisons with standard designs. Seven scenarios, which
represent the gamut of probable doseresponse curves were chosen as shown in panel a in the figure.
In simulations, it was found that across all seven scenarios, a single adaptive trial can replace two
standard trials (PoC and dose-ranging). The power of the trend test for PoC was always greater for the
adaptive design, as shown in panel b. When there was a small doseresponse effect (scenarios 2
and 3), the power of the adaptive design was about double that of the standard design. When the effect
size was modest (scenarios 4 and 5), the power was increased to practically 100%. When effect sizes
were large (scenarios 6 and 7), the power was almost 100% for both adaptive and standard designs.
For the same total sample size, the adaptive combined PoCdose-finding trial is more efficient
than the two standard trials in estimating the response at every dose (see panel c). The continuous
curve shows the efficiency of the adaptive design relative to the standard dose-ranging design for
scenario 7. Efficiency at each dose is defined as the ratio of the square of the estimation error of the
standard design to the square of the estimation error of the adaptive design. The bars show the
number of subjects allocated to each dose by the adaptive design. These results are computed by
averaging the results of 1,000 simulations.The overall efficiency across all doses is greater by a
factor of five, whereas for the sloping part of the dose response curve (doses 4, 5 and 6) the adaptive
design is three times more efficient. In panel d, the adaptive combined PoCdose-ranging trial with
60 subjects is as efficient in estimating the response at every dose as the two standard trials with a
combined sample size of 120 subjects. It is also as powerful in testing for PoC.
These results are true irrespective of which of the seven scenarios reflects the true doseresponse
curve. For all seven scenarios for the same sample size, the efficiency of the adaptive design was
about five times that of the standard design over all doses. It was three times that of the standard
design for estimating doseresponse in the sloping part of the doseresponse curve. Another way to
think about this result is that for half the sample size, the adaptive design is as powerful and efficient
as the standard approach with two trials.
PersPecti ves
L
e
v
e
l
280 336 392 448 504 560 224 168 112 56
Time (days)
56 0
Total IL-1 (complex)
Antibody
Free IL-1 is suppressed
C-reactive protein
Symptoms
Flaring
Remission
selected group(s) with the control will use
data from the continuing groups from both
stages of the trial.
There are three key potential advantages
of seamless adaptive designs: a reduction
in the duration of the clinical development
programme, by eliminating the time lag that
traditionally occurs between Phase II and
III trials; greater efficiency from the use of
data from both stages, which might mean
that fewer patients are required to obtain
the same quality of information; and earlier
acquisition of long-term safety data, gathered
through continued follow-up of patients
from the first stage (see Supplementary
information S3 (figure))
25,29
.
not all drug development programmes
will be candidates for these designs.
Feasibility considerations for use of these
designs include the length of follow-up
time for the endpoint used for selection
compared with duration of enrolment.
Shorter follow-up will be more conducive
to a seamless adaptive design, whereas a
relatively long endpoint follow-up period
will tend to militate against using such a
design. development programmes that do
not involve complex treatment regimens
might therefore be better suited to such
designs. drug supply and drug packaging
will be expected to be more challenging in
this setting.
a number of logistical and regulatory
actions must be fulfilled to avoid compro-
mising an adaptive trial. First, the actual
algorithm for determining the adaptation
to implement must be specified in advance.
This is usually accomplished by creating a
charter for the independent data monitoring
committee charged with the responsibility
of performing the unblinded interim analysis
and communicating as appropriate with
the sponsor. In addition, the sponsor must
have developed in-house procedures to
ensure that the algorithm is not transmitted
throughout the company, and especially not
to the study investigators.
To maintain trial integrity, the processes
by which interim data are examined and
selection decisions are made and imple-
mented must be considered very carefully.
Current conventions that restrict knowledge
of interim results in ongoing trials should
be respected to avoid compromising the
interpretability of trial results. In some cases
the decision being made at the selection
point of a seamless design will be one for
which sponsor perspective might be relevant
and which has traditionally been a sponsor
responsibility, raising the question of sponsor
involvement in the monitoring process.
a distinction is sometimes made between
seamless adaptive designs that are inferen-
tially seamless or operationally seamless.
In inferentially seamless designs, which we
describe here, the main analysis uses data
from both stages of the trial. In operation-
ally seamless designs, the final analysis only
uses data from patients enrolled after the
selection decision. This may allow a broader
investigation of the first-stage data involving
sponsor personnel and decreases concerns
about trial integrity; in addition, traditional
non-adaptive statistical methodology nor-
mally suffices. Such designs may maintain
the advantage of reducing white space, while
losing the efficiency that results from using
data accrued in stages. regardless, operating
procedures for the monitoring process in
seamless designs must be carefully considered
to ensure that the right expertise is applied
to the decision, while limiting access to the
accruing data as appropriate to maintain
trial integrity.
other considerations for adaptive designs
include the endpoint used for selection.
This need not be the same as the endpoint to
be used in the main study analysis; if a good
surrogate marker is available, this can be
used and might enhance the efficiency of the
seamless trial. Second, modelling and simula-
tion will probably have a very important role
in developing the specific details of seamless
designs (for example, per-group sample sizes
in the different stages, considered under vari-
ous scenarios) to ensure that they are robust
and efficient. Third, the final analysis must
use statistical methodology that is appropriate
for the design: naive comparisons of control
Figure 2 | Dose selection in the development of a therapeutic for MuckleWells syndrome.
MuckleWells syndrome is a rare genetic disorder characterized by fever, urticaria, joint pain and
malaise. A monoclonal antibody against interleukin-1 (iL-1 ), canakinumab, has been developed to
treat this iL-1-dependent inflammatory disease. The antibody is delivered parenterally and binds to
free iL-1, driving it into the inactive complex and leading to remission of symptoms
21
. Total iL-1,
which represents mainly the inactive complex, increases after dosing and can be measured. By the
laws of mass action, the free and active form of iL-1, which cannot be measured, must decrease.
However, the reduction in free iL-1 results in a decrease in markers of inflammation, including
c-reactive protein (which can be measured), and a remission of clinical signs and symptoms of disease.
The clinical data on these relationships can be captured in a mathematical model, shown in the figure,
which is continuously adjusted in the light of new data. This framework simulation could then be used
to propose a suitable dose and dosing regimen that would be predicted to produce a desired response
for the majority of patients (for example, an 80% probability that 90% of patients will be flare-free for
2 months).
PersPecti ves
M
e
a
n

r
e
s
p
o
n
s
e
25
20
15
10
5
0
6 8 7 5 4 3 2
Dose
Scenario 1
Scenario 2
Scenario 3
1
a
Scenario 4
Scenario 5
Scenario 6
Scenario 7
A
v
e
r
a
g
e

a
l
l
o
c
a
t
i
o
n
20
18
16
12
14
10
8
6
4
2
0
Dose
Average allocation
Efficiency
c
b
E
f
f
i
c
i
e
n
c
y
E
f
f
i
c
i
e
n
c
y
10
9
8
6
7
5
4
3
2
1
0
10
9
8
6
7
5
4
3
2
1
0
6 7 5 4 3 2
Dose
Standard design (sample size = 120)
1
d
1 2 3 4 5 6 7
Scenario Power of
standard
PoC trial*
Power of
combined
adaptive design
1
2
3
4
5
6
7
5
33
32
86
85
100
99
5
84
67
100
99
100
100
*Sample size = 30.
versus the selected treatment that do not
account for the design will not be appropriate.
Finally, the appropriateness of the design
does not depend on any particular algorithm
for choosing the patient group to be contin-
ued; it is not even necessary for a firm
algorithm to be specified in advance, although
the general principles that will govern the
decision should be clear in advance.
Sample size re-estimation within a
confirmatory trial (Phase III). Sample size
re-estimation (SSr) provides a mechanism for
appropriately using the information obtained
during a confirmatory study to inform and
adjust the necessary sample size going for-
ward
30,31
. This process increases confidence
that an appropriate sample size has been
chosen to answer the primary study questions.
The standard approach used to power a
confirmatory study is to first estimate the
underlying treatment effect on the primary
endpoint based on available prior informa-
tion. The parameter denotes the true
underlying difference between the treat-
ment and control arms with respect to the
primary endpoint. even though the true
value of is unknown, the trial investigators
will usually have in mind a specific value,
min
, which represents the smallest clinically
important delta (SCId) for this clinical trial.
next, the trial designers will determine the
sample size that can detect values of , based
on prior information, that exceed the SCId
with good power. The standard deviation
(between subject variability) is a nuisance
parameter whose true value must be
estimated in order to proceed with the
sample size calculation.
The SCId can often be pre-specified
from purely clinical arguments, whereas the
actual effect size is unknown. Therefore, it is
possible in principle to design a study with
a fixed sample size that will have adequate
power to detect the SCId, in the absence
of adequate prior information about the
actual effect size of the test agent. This is
what statisticians envisaged when they
created the fixed-sample methodology.
However, this fixed sample methodology
has several drawbacks. If the actual effect is
substantially larger than the SCId, a smaller
sample size would have sufficed to attain
adequate power
32
.
Sponsors will not often risk significant
resources on trial sizes based on SCId
assumptions that would lead to larger trials
than the current best guess about the actual
effect size (BOX 4). Instead, a smaller trial
corresponding to that best guess may be run;
if that assumption is too optimistic, and the
Box 2 | Case study: combining poC and dose-ranging trials into a single adaptive trial
This example illustrates how a single adaptive trial can replace two standard trials
proof-of-concept (PoC) and dose-ranging and that the combined trial has greater power than
the standard PoC design, and is substantially better at estimating the doseresponse curve.
The trial evaluated an analgesic drug to treat dental pain and tested seven doses of the drug.
Several designs with different sample sizes, randomization ratios of drug to placebo and starting
doses were simulated against several scenarios. Here, we describe one design with a sample
size of 120 subjects (40 placebo, 80 drug). Bayesian adaptive trials were simulated over seven
drugresponse scenarios to enable comparisons with standard designs. Seven scenarios, which
represent the gamut of probable doseresponse curves were chosen as shown in panel a in the figure.
In simulations, it was found that across all seven scenarios, a single adaptive trial can replace two
standard trials (PoC and dose-ranging). The power of the trend test for PoC was always greater for the
adaptive design, as shown in panel b. When there was a small doseresponse effect (scenarios 2
and 3), the power of the adaptive design was about double that of the standard design. When the effect
size was modest (scenarios 4 and 5), the power was increased to practically 100%. When effect sizes
were large (scenarios 6 and 7), the power was almost 100% for both adaptive and standard designs.
For the same total sample size, the adaptive combined PoCdose-finding trial is more efficient
than the two standard trials in estimating the response at every dose (see panel c). The continuous
curve shows the efficiency of the adaptive design relative to the standard dose-ranging design for
scenario 7. Efficiency at each dose is defined as the ratio of the square of the estimation error of the
standard design to the square of the estimation error of the adaptive design. The bars show the
number of subjects allocated to each dose by the adaptive design. These results are computed by
averaging the results of 1,000 simulations.The overall efficiency across all doses is greater by a
factor of five, whereas for the sloping part of the dose response curve (doses 4, 5 and 6) the adaptive
design is three times more efficient. In panel d, the adaptive combined PoCdose-ranging trial with
60 subjects is as efficient in estimating the response at every dose as the two standard trials with a
combined sample size of 120 subjects. It is also as powerful in testing for PoC.
These results are true irrespective of which of the seven scenarios reflects the true doseresponse
curve. For all seven scenarios for the same sample size, the efficiency of the adaptive design was
about five times that of the standard design over all doses. It was three times that of the standard
design for estimating doseresponse in the sloping part of the doseresponse curve. Another way to
think about this result is that for half the sample size, the adaptive design is as powerful and efficient
as the standard approach with two trials.
PersPecti ves
L
e
v
e
l
280 336 392 448 504 560 224 168 112 56
Time (days)
56 0
Total IL-1 (complex)
Antibody
Free IL-1 is suppressed
C-reactive protein
Symptoms
Flaring
Remission
selected group(s) with the control will use
data from the continuing groups from both
stages of the trial.
There are three key potential advantages
of seamless adaptive designs: a reduction
in the duration of the clinical development
programme, by eliminating the time lag that
traditionally occurs between Phase II and
III trials; greater efficiency from the use of
data from both stages, which might mean
that fewer patients are required to obtain
the same quality of information; and earlier
acquisition of long-term safety data, gathered
through continued follow-up of patients
from the first stage (see Supplementary
information S3 (figure))
25,29
.
not all drug development programmes
will be candidates for these designs.
Feasibility considerations for use of these
designs include the length of follow-up
time for the endpoint used for selection
compared with duration of enrolment.
Shorter follow-up will be more conducive
to a seamless adaptive design, whereas a
relatively long endpoint follow-up period
will tend to militate against using such a
design. development programmes that do
not involve complex treatment regimens
might therefore be better suited to such
designs. drug supply and drug packaging
will be expected to be more challenging in
this setting.
a number of logistical and regulatory
actions must be fulfilled to avoid compro-
mising an adaptive trial. First, the actual
algorithm for determining the adaptation
to implement must be specified in advance.
This is usually accomplished by creating a
charter for the independent data monitoring
committee charged with the responsibility
of performing the unblinded interim analysis
and communicating as appropriate with
the sponsor. In addition, the sponsor must
have developed in-house procedures to
ensure that the algorithm is not transmitted
throughout the company, and especially not
to the study investigators.
To maintain trial integrity, the processes
by which interim data are examined and
selection decisions are made and imple-
mented must be considered very carefully.
Current conventions that restrict knowledge
of interim results in ongoing trials should
be respected to avoid compromising the
interpretability of trial results. In some cases
the decision being made at the selection
point of a seamless design will be one for
which sponsor perspective might be relevant
and which has traditionally been a sponsor
responsibility, raising the question of sponsor
involvement in the monitoring process.
a distinction is sometimes made between
seamless adaptive designs that are inferen-
tially seamless or operationally seamless.
In inferentially seamless designs, which we
describe here, the main analysis uses data
from both stages of the trial. In operation-
ally seamless designs, the final analysis only
uses data from patients enrolled after the
selection decision. This may allow a broader
investigation of the first-stage data involving
sponsor personnel and decreases concerns
about trial integrity; in addition, traditional
non-adaptive statistical methodology nor-
mally suffices. Such designs may maintain
the advantage of reducing white space, while
losing the efficiency that results from using
data accrued in stages. regardless, operating
procedures for the monitoring process in
seamless designs must be carefully considered
to ensure that the right expertise is applied
to the decision, while limiting access to the
accruing data as appropriate to maintain
trial integrity.
other considerations for adaptive designs
include the endpoint used for selection.
This need not be the same as the endpoint to
be used in the main study analysis; if a good
surrogate marker is available, this can be
used and might enhance the efficiency of the
seamless trial. Second, modelling and simula-
tion will probably have a very important role
in developing the specific details of seamless
designs (for example, per-group sample sizes
in the different stages, considered under vari-
ous scenarios) to ensure that they are robust
and efficient. Third, the final analysis must
use statistical methodology that is appropriate
for the design: naive comparisons of control
Figure 2 | Dose selection in the development of a therapeutic for MuckleWells syndrome.
MuckleWells syndrome is a rare genetic disorder characterized by fever, urticaria, joint pain and
malaise. A monoclonal antibody against interleukin-1 (iL-1 ), canakinumab, has been developed to
treat this iL-1-dependent inflammatory disease. The antibody is delivered parenterally and binds to
free iL-1, driving it into the inactive complex and leading to remission of symptoms
21
. Total iL-1,
which represents mainly the inactive complex, increases after dosing and can be measured. By the
laws of mass action, the free and active form of iL-1, which cannot be measured, must decrease.
However, the reduction in free iL-1 results in a decrease in markers of inflammation, including
c-reactive protein (which can be measured), and a remission of clinical signs and symptoms of disease.
The clinical data on these relationships can be captured in a mathematical model, shown in the figure,
which is continuously adjusted in the light of new data. This framework simulation could then be used
to propose a suitable dose and dosing regimen that would be predicted to produce a desired response
for the majority of patients (for example, an 80% probability that 90% of patients will be flare-free for
2 months).
PersPecti ves

Box 5 | group-sequential and adaptive designs for sample size re-estimation
Group-sequential design. Suppose that the sponsor is unsure of the true value of , but nevertheless
believes that it is larger than the smallest clinically important delta (SCID). In this case, a group-
sequential design might be considered. Such a design is characterized by a maximum sample size,
an interim monitoring strategy and a corresponding boundary for early stopping for efficacy.
The maximum sample size is computed so that the study has adequate power to detect a value
of that the sponsor believes represents a reasonable estimate of the efficacy of the experimental
compound, provided this estimate is at least as large as the SCID. If the sponsor wishes to be very
conservative about this estimate, the maximum sample size needed can be computed to have
adequate power at the SCID itself. An up-front commitment is made to enrol patients up to this
maximum sample size. However, if the true exceeds the SCID, the trial may terminate earlier
with high probability by crossing an early stopping boundary at an interim analysis.
Returning to the example discussed in BOX 4, suppose that the sponsor decides to make an
up-front commitment of 4,000 patients to the trial but intends to monitor the accruing data up to
four times, after 1,000, 2,000, 3,000 and 4,000 patients become evaluable for the primary endpoint.
The commitment of 4,000 patients ensures that the trial will have 88% power to detect a
difference as small as = 0.1 (in this case the SCID). Although this is a rather large sample size to
commit to the trial, the actual sample size is expected to be substantially smaller if the true is
larger than the SCID. This is because at each of the four interim monitoring time points there is a
chance of early termination and a declaration of statistical significance. At each interim analysis,
a test for statistical significance using all available primary endpoint data would be performed,
and the result would be compared with a properly determined early-stopping boundary value.
The trial could be terminated the first time that a boundary is reached, with a valid claim that the
experimental arm is more efficacious than the control arm.
However, sometimes a sponsor might not be willing to make such a large up-front commitment,
particularly when the only currently available data on are from one or two small Phase II trials.
The sponsor might feel more comfortable with a design that starts out with a smaller sample size
of, say, 1,000 patients, with the opportunity to increase the sample size at an interim time point
and after observing data from the trial. This is the motivation for the adaptive design below.
The adaptive design. The group-sequential design described above is characterized by
pre-specifying a maximum sample size up-front and terminating earlier if the true is larger than
anticipated. By contrast, an adaptive design pre-specifies a smaller initial sample size, but with
the possibility of increasing the commitment after seeing some interim data from the trial. On the
surface, this is similar to the usual practice of first running a small Phase II trial to obtain an idea
about efficacy and safety and then following it up with a larger Phase III trial once the efficacy and
safety of the compound have been established. There is, however, an important distinction between
the conventional Phase II followed by Phase III strategy and the adaptive strategy outlined below.
In the conventional approach, the data from the Phase II trial are not combined with the data
from the Phase III trial. The adaptive design, however, uses all the data from both stages for the
final analysis. This can have important advantages both in terms of gaining additional statistical
power, as well as shortening the drug development time. In our example, we stated that the SCID
was 0.1. Supposing that the sponsor believes that the true = 0.2 that is, twice as large as the
SCID if this is indeed the case, then a total sample size of 1,000 patients will have 89% power at
a one-sided level of 0.025. On this basis, the sponsor is prepared to make an initial investment of
1,000 patients to this trial. As an insurance policy, however, the sponsor intends to take an interim
look at the accruing data at the mid-point of the trial, after 500 patients are evaluable for
response. If the estimate of obtained from these 500 is smaller than the sponsor expected,
then the sponsor might choose to increase the sample size to preserve the power of the trial.
Many different criteria can be used to decide whether an increase in sample size is warranted.
A commonly used criterion is conditional power. The conditional power at an interim analysis is
the probability, given the observed data, that the experimental compound will demonstrate
efficacy on completion of the trial. The conditional power computation requires specifying a
value for . One can choose the value specified at the initial design stage or the value estimated
from the interim data. In this example, we use the interim estimated value of for evaluating
conditional power. The table below displays conditional power for various estimated values of
at the interim time point, along with the total sample size needed to achieve 80% conditional
power at the final analysis. The entries in the table assume that = 1.
interim
estimate ()
conditional power without
sample size increase
total sample size needed to achieve
80% conditional power
0.2 95% 720 (sample size reduction)
0.15 72% 1,166 (sample size increase)
how the new sample size will be calculated.
Usually, this type of adjustment will not be
permitted by regulatory authorities more
than once.
For unblinded sample size re-estimation,
the sponsor sets up a mechanism (possibly
with the data monitoring committee of
the trial) whereby the SSr is based on an
unblinded estimate of variability (or statistical
information) at the interim analysis. Sample
size may be altered one or more times,
but the maximum statistical information
must be pre-specified.
If the sponsor agrees that there will
be no early stopping for efficacy following
an interim analysis, then no adjustment to
the final analysis criteria is necessary.
The data monitoring committee may
monitor the data one or more times and
adjust the sample size up or down based
on the unblinded estimate of variability
and attempt to reach the pre-specified
maximum information.
When the sponsor pre-specifies the
interim time points at which it is permis-
sible to terminate early for efficacy, the
criteria for each interim analysis must be
pre-specified in a manner that controls the
false-positive rate across the entire study.
This will result in adjustment to the final
analysis criterion if the study is not stopped
early. Interim looks undertaken solely for
administrative purposes, with no inten-
tion of stopping the trial in light of efficacy
data, do not need to have defined criteria.
The trial then proceeds until either it is ter-
minated early for efficacy on the basis of the
pre-defined criteria being reached, or until
the planned maximum information (sample
size or number of events) is reached.
Tackling challenges of new trial designs
Because they are so flexible, these new trial
designs require significant statistical analyses,
simulations and logistical considerations
to verify their operating characteristics,
and therefore tend to require more time
for the planning and protocol development
phase. regulatory agencies and Institutional
review Boards also need to approve the
design format for interim analysis, and these
discussions can sometimes take considerable
time. Such time considerations can lead a
company to follow the traditional route to
clinical development, without fully appre-
ciating the advantages that adaptive designs
can eventually bring in terms of time and
cost savings, and probability of success.
as described above, adaptive designs
further require the following: quickly
observable responses relative to the patient
PersPecti ves
R
e
s
p
o
n
s
e
Wasted doses Dose Wasted doses
Box 3 | Adaptive dose finding
In an adaptive dose-finding study, the dose assignment(s) to the next subject, or next cohort of
patients, is based on responses of previous subjects, and the dose assignment is chosen to maximize
the information about the doseresponse curve, according to some pre-defined objective metric
(for example, variability in parameter estimates). In a traditional dose-finding trial, selecting a few
doses may not adequately represent the doseresponse relationship and many patients will be
allocated to non-informative doses (wasted doses), as shown in the figure. In adaptive dose-finding,
the strategy is to initially include only a few patients on many doses to explore the doseresponse,
then to allocate the dose range of interest to more patients. This reduces the allocation of patients
to non-informative doses
27,28
. Compared with fixed randomization, this approach has the ethical
advantage that fewer subjects are assigned doses that are too high or too low. It can also avoid
additional, separate trials that might be necessary when fixed dose-finding trials do not adequately
define the dose range.
Adaptive dose-finding trials also require
an infrastructure that allows the rapid
communication of responses from trial sites to a
central unblinded analysis centre and of adaptive
dose assignments to the trial sites. Randomization
software capable of rapidly computing dynamic
allocation of doses to subjects is additionally
mandated by adaptive trials because pre-specified
randomization lists will not work. In addition,
a flexible drug-supply process is required because
demand for doses is not fixed in advance,
but rather evolves as information on responses at
various doses is gathered as the trial progresses.

Box 4 | issues with the standard clinical development approach
Issues with the standard approach to clinical development can be illustrated by considering a
randomized clinical trial with the following assumptions. Based on available evidence from
early-phase trials, it is estimated that = 1, that the anticipated effect size = 0.2 and that the
smallest clinically important delta (SCID) is 0.1. Rather than conservatively enrolling a sample size
required to demonstrate the SCID (4,000 subjects), the sponsor appropriately powers the trial to
detect the estimated larger (1,000 subjects). Now, suppose that the true underlying value of is
0.15. In that case, a sample size of 2,000 subjects would be required to adequately power the trial
to detect this difference. The difficulty is, of course, that the true underlying value of is not
known at the start of the trial. In this example, the 1,000-subject study would probably yield a
non-significant result, as it is only powered to detect an effect size of 0.2, which is larger than the
actual effect size of 0.15.
In this example, unless the 1,000-patient under-powered trial was repeated with a larger sample
size, then a potentially efficacious treatment would be unnecessarily and unfortunately discarded.
If the trial were to be repeated with the re-estimation of the actual effect size, then 2,000 patients
would need to be enrolled, and the time and resources to perform the original trial (sometimes
more than 3 years) would have been spent without much benefit other than gaining a more
reliable estimate of the actual effect size in order to design the second trial. More importantly,
the subjects for that study would have been put at unnecessary risk because the study had no real
chance of being definitive.
truth is an effect size closer to the SCId,
the trial will be underpowered and therefore
have a high chance of failure.
one approach to solving the problem of
uncertainty about is to design and execute
an additional number of exploratory trials
(typically Phase II studies). These small
Phase II studies are normally carried out to
get a more precise estimate (or best guess)
of the actual and so that the confirma-
tory study might be adequately powered.
each exploratory trial, although somewhat
smaller than confirmatory trials, still
requires significant resources to perform
appropriately. also, the inevitable start-up
time and wind-down activities between
trials have to be included when determining
true programme efficiency and develop-
ment timelines. This might therefore not
be the most efficient way to proceed from
the viewpoint of the entire clinical trial
programme.
Advantages of adaptive SSR in confirmatory
trials. a more flexible approach to the
fixed sample-size methodology is needed.
By altering the sample size using interim
data from the trial itself, this flexibility can
be achieved without compromising the
power or the false-positive rate of the trial
(that is, the chance of making a false claim
of efficacy for a treatment that is not effica-
cious). SSr should be considered in two
situations: when there is significant uncer-
tainty about ; or when there is a substantial
difference between the sample size resulting
from using the SCId and the sample size the
sponsor can justify on the basis of their best
guess of the effect size
29
.
SSr usually involves the choice of a suitable
initial sample size, including one or more
interim analyses at which the sample size
will be re-assessed
30
. There are two distinct
strategies the group sequential strategy
and the adaptive SSr strategy for choosing
the initial sample size, and then altering it on
the basis of data obtained at various interim
analysis time points. The group sequential
strategy, which is also an adaptive design,
begins with a large up-front sample size
commitment and cuts back if the accruing
data suggest that the large sample size is not
needed. The adaptive SSr strategy proceeds
in the opposite direction, starting out with
a smaller initial sample size commitment
but with the option to increase it should the
accruing data suggest that such an increase is
warranted
3033
(BOX 5).
Extending the methodology to unknown .
although the group sequential and adap-
tive SSr methods were presented under the
assumption that the standard deviation
is known, they apply equally for the case
of unknown
3032
. one can start out with
an initial estimate of and corresponding
sample-size estimate. Then, following an
interim analysis, one can re-estimate this
nuisance parameter, input the updated esti-
mate into the equation and re-compute the
sample size. an illustrative example is given
in FIG. 3.
There are two ways to obtain the new
sample size in the situation of unknown :
blinded and unblinded. In the instance of
blinded sample size re-estimation, the
sponsor uses pooled data to estimate .
This is permitted with no penalty to the
analysis criteria (that is, alpha, or the
probability of Type I (false positive) error).
It is preferable that the sponsor pre-specifies
how many times changes are to be made
to the sample size, at what time points and
PersPecti ves

Box 5 | group-sequential and adaptive designs for sample size re-estimation
Group-sequential design. Suppose that the sponsor is unsure of the true value of , but nevertheless
believes that it is larger than the smallest clinically important delta (SCID). In this case, a group-
sequential design might be considered. Such a design is characterized by a maximum sample size,
an interim monitoring strategy and a corresponding boundary for early stopping for efficacy.
The maximum sample size is computed so that the study has adequate power to detect a value
of that the sponsor believes represents a reasonable estimate of the efficacy of the experimental
compound, provided this estimate is at least as large as the SCID. If the sponsor wishes to be very
conservative about this estimate, the maximum sample size needed can be computed to have
adequate power at the SCID itself. An up-front commitment is made to enrol patients up to this
maximum sample size. However, if the true exceeds the SCID, the trial may terminate earlier
with high probability by crossing an early stopping boundary at an interim analysis.
Returning to the example discussed in BOX 4, suppose that the sponsor decides to make an
up-front commitment of 4,000 patients to the trial but intends to monitor the accruing data up to
four times, after 1,000, 2,000, 3,000 and 4,000 patients become evaluable for the primary endpoint.
The commitment of 4,000 patients ensures that the trial will have 88% power to detect a
difference as small as = 0.1 (in this case the SCID). Although this is a rather large sample size to
commit to the trial, the actual sample size is expected to be substantially smaller if the true is
larger than the SCID. This is because at each of the four interim monitoring time points there is a
chance of early termination and a declaration of statistical significance. At each interim analysis,
a test for statistical significance using all available primary endpoint data would be performed,
and the result would be compared with a properly determined early-stopping boundary value.
The trial could be terminated the first time that a boundary is reached, with a valid claim that the
experimental arm is more efficacious than the control arm.
However, sometimes a sponsor might not be willing to make such a large up-front commitment,
particularly when the only currently available data on are from one or two small Phase II trials.
The sponsor might feel more comfortable with a design that starts out with a smaller sample size
of, say, 1,000 patients, with the opportunity to increase the sample size at an interim time point
and after observing data from the trial. This is the motivation for the adaptive design below.
The adaptive design. The group-sequential design described above is characterized by
pre-specifying a maximum sample size up-front and terminating earlier if the true is larger than
anticipated. By contrast, an adaptive design pre-specifies a smaller initial sample size, but with
the possibility of increasing the commitment after seeing some interim data from the trial. On the
surface, this is similar to the usual practice of first running a small Phase II trial to obtain an idea
about efficacy and safety and then following it up with a larger Phase III trial once the efficacy and
safety of the compound have been established. There is, however, an important distinction between
the conventional Phase II followed by Phase III strategy and the adaptive strategy outlined below.
In the conventional approach, the data from the Phase II trial are not combined with the data
from the Phase III trial. The adaptive design, however, uses all the data from both stages for the
final analysis. This can have important advantages both in terms of gaining additional statistical
power, as well as shortening the drug development time. In our example, we stated that the SCID
was 0.1. Supposing that the sponsor believes that the true = 0.2 that is, twice as large as the
SCID if this is indeed the case, then a total sample size of 1,000 patients will have 89% power at
a one-sided level of 0.025. On this basis, the sponsor is prepared to make an initial investment of
1,000 patients to this trial. As an insurance policy, however, the sponsor intends to take an interim
look at the accruing data at the mid-point of the trial, after 500 patients are evaluable for
response. If the estimate of obtained from these 500 is smaller than the sponsor expected,
then the sponsor might choose to increase the sample size to preserve the power of the trial.
Many different criteria can be used to decide whether an increase in sample size is warranted.
A commonly used criterion is conditional power. The conditional power at an interim analysis is
the probability, given the observed data, that the experimental compound will demonstrate
efficacy on completion of the trial. The conditional power computation requires specifying a
value for . One can choose the value specified at the initial design stage or the value estimated
from the interim data. In this example, we use the interim estimated value of for evaluating
conditional power. The table below displays conditional power for various estimated values of
at the interim time point, along with the total sample size needed to achieve 80% conditional
power at the final analysis. The entries in the table assume that = 1.
interim
estimate ()
conditional power without
sample size increase
total sample size needed to achieve
80% conditional power
how the new sample size will be calculated.
Usually, this type of adjustment will not be
permitted by regulatory authorities more
than once.
For unblinded sample size re-estimation,
the sponsor sets up a mechanism (possibly
with the data monitoring committee of
the trial) whereby the SSr is based on an
unblinded estimate of variability (or statistical
information) at the interim analysis. Sample
size may be altered one or more times,
but the maximum statistical information
must be pre-specified.
If the sponsor agrees that there will
be no early stopping for efficacy following
an interim analysis, then no adjustment to
the final analysis criteria is necessary.
The data monitoring committee may
monitor the data one or more times and
adjust the sample size up or down based
on the unblinded estimate of variability
and attempt to reach the pre-specified
maximum information.
When the sponsor pre-specifies the
interim time points at which it is permis-
sible to terminate early for efficacy, the
criteria for each interim analysis must be
pre-specified in a manner that controls the
false-positive rate across the entire study.
This will result in adjustment to the final
analysis criterion if the study is not stopped
early. Interim looks undertaken solely for
administrative purposes, with no inten-
tion of stopping the trial in light of efficacy
data, do not need to have defined criteria.
The trial then proceeds until either it is ter-
minated early for efficacy on the basis of the
pre-defined criteria being reached, or until
the planned maximum information (sample
size or number of events) is reached.
Tackling challenges of new trial designs
Because they are so flexible, these new trial
designs require significant statistical analyses,
simulations and logistical considerations
to verify their operating characteristics,
and therefore tend to require more time
for the planning and protocol development
phase. regulatory agencies and Institutional
review Boards also need to approve the
design format for interim analysis, and these
discussions can sometimes take considerable
time. Such time considerations can lead a
company to follow the traditional route to
clinical development, without fully appre-
ciating the advantages that adaptive designs
can eventually bring in terms of time and
cost savings, and probability of success.
as described above, adaptive designs
further require the following: quickly
observable responses relative to the patient
PersPecti ves
R
e
s
p
o
n
s
e
Wasted doses Dose Wasted doses
Box 3 | Adaptive dose finding
In an adaptive dose-finding study, the dose assignment(s) to the next subject, or next cohort of
patients, is based on responses of previous subjects, and the dose assignment is chosen to maximize
the information about the doseresponse curve, according to some pre-defined objective metric
(for example, variability in parameter estimates). In a traditional dose-finding trial, selecting a few
doses may not adequately represent the doseresponse relationship and many patients will be
allocated to non-informative doses (wasted doses), as shown in the figure. In adaptive dose-finding,
the strategy is to initially include only a few patients on many doses to explore the doseresponse,
then to allocate the dose range of interest to more patients. This reduces the allocation of patients
to non-informative doses
27,28
. Compared with fixed randomization, this approach has the ethical
advantage that fewer subjects are assigned doses that are too high or too low. It can also avoid
additional, separate trials that might be necessary when fixed dose-finding trials do not adequately
define the dose range.
Adaptive dose-finding trials also require
an infrastructure that allows the rapid
communication of responses from trial sites to a
central unblinded analysis centre and of adaptive
dose assignments to the trial sites. Randomization
software capable of rapidly computing dynamic
allocation of doses to subjects is additionally
mandated by adaptive trials because pre-specified
randomization lists will not work. In addition,
a flexible drug-supply process is required because
demand for doses is not fixed in advance,
but rather evolves as information on responses at
various doses is gathered as the trial progresses.

Box 4 | issues with the standard clinical development approach
Issues with the standard approach to clinical development can be illustrated by considering a
randomized clinical trial with the following assumptions. Based on available evidence from
early-phase trials, it is estimated that = 1, that the anticipated effect size = 0.2 and that the
smallest clinically important delta (SCID) is 0.1. Rather than conservatively enrolling a sample size
required to demonstrate the SCID (4,000 subjects), the sponsor appropriately powers the trial to
detect the estimated larger (1,000 subjects). Now, suppose that the true underlying value of is
0.15. In that case, a sample size of 2,000 subjects would be required to adequately power the trial
to detect this difference. The difficulty is, of course, that the true underlying value of is not
known at the start of the trial. In this example, the 1,000-subject study would probably yield a
non-significant result, as it is only powered to detect an effect size of 0.2, which is larger than the
actual effect size of 0.15.
In this example, unless the 1,000-patient under-powered trial was repeated with a larger sample
size, then a potentially efficacious treatment would be unnecessarily and unfortunately discarded.
If the trial were to be repeated with the re-estimation of the actual effect size, then 2,000 patients
would need to be enrolled, and the time and resources to perform the original trial (sometimes
more than 3 years) would have been spent without much benefit other than gaining a more
reliable estimate of the actual effect size in order to design the second trial. More importantly,
the subjects for that study would have been put at unnecessary risk because the study had no real
chance of being definitive.
truth is an effect size closer to the SCId,
the trial will be underpowered and therefore
have a high chance of failure.
one approach to solving the problem of
uncertainty about is to design and execute
an additional number of exploratory trials
(typically Phase II studies). These small
Phase II studies are normally carried out to
get a more precise estimate (or best guess)
of the actual and so that the confirma-
tory study might be adequately powered.
each exploratory trial, although somewhat
smaller than confirmatory trials, still
requires significant resources to perform
appropriately. also, the inevitable start-up
time and wind-down activities between
trials have to be included when determining
true programme efficiency and develop-
ment timelines. This might therefore not
be the most efficient way to proceed from
the viewpoint of the entire clinical trial
programme.
Advantages of adaptive SSR in confirmatory
trials. a more flexible approach to the
fixed sample-size methodology is needed.
By altering the sample size using interim
data from the trial itself, this flexibility can
be achieved without compromising the
power or the false-positive rate of the trial
(that is, the chance of making a false claim
of efficacy for a treatment that is not effica-
cious). SSr should be considered in two
situations: when there is significant uncer-
tainty about ; or when there is a substantial
difference between the sample size resulting
from using the SCId and the sample size the
sponsor can justify on the basis of their best
guess of the effect size
29
.
SSr usually involves the choice of a suitable
initial sample size, including one or more
interim analyses at which the sample size
will be re-assessed
30
. There are two distinct
strategies the group sequential strategy
and the adaptive SSr strategy for choosing
the initial sample size, and then altering it on
the basis of data obtained at various interim
analysis time points. The group sequential
strategy, which is also an adaptive design,
begins with a large up-front sample size
commitment and cuts back if the accruing
data suggest that the large sample size is not
needed. The adaptive SSr strategy proceeds
in the opposite direction, starting out with
a smaller initial sample size commitment
but with the option to increase it should the
accruing data suggest that such an increase is
warranted
3033
(BOX 5).
Extending the methodology to unknown .
although the group sequential and adap-
tive SSr methods were presented under the
assumption that the standard deviation
is known, they apply equally for the case
of unknown
3032
. one can start out with
an initial estimate of and corresponding
sample-size estimate. Then, following an
interim analysis, one can re-estimate this
nuisance parameter, input the updated esti-
mate into the equation and re-compute the
sample size. an illustrative example is given
in FIG. 3.
There are two ways to obtain the new
sample size in the situation of unknown :
blinded and unblinded. In the instance of
blinded sample size re-estimation, the
sponsor uses pooled data to estimate .
This is permitted with no penalty to the
analysis criteria (that is, alpha, or the
probability of Type I (false positive) error).
It is preferable that the sponsor pre-specifies
how many times changes are to be made
to the sample size, at what time points and
PersPecti ves
26. Gallo, P. et al. Adaptive designs in clinical drug
development an executive summary of the PhRMA
Working Group. J. Biopharm. Stat. 16, 275283
(2006).
27. Bornkamp, B. et al. Innovative approaches for
designing and analyzing adaptive dose-ranging trials.
J. Biopharm. Stat. 17, 965995 (2007).
28. Weir, C. J., Spiegelhalter, D. J. & Grieve, A. P.
Flexible design and efficient implementation of
adaptive dose-finding studies. J. Biopharm. Stat.
17, 10331050 (2007).
29. Bretz, F., Schmidli, H., Knig, F., Racine, A. &
Maurer, W. Confirmatory seamless Phase II/III clinical
trials with hypotheses selection at interim: general
concepts. Biom. J. 48, 623634 (2006).
30. Mehta, C. R. & Patel, N. R. Adaptive, group
sequential and decision theoretic approaches
to sample size determination. Stat. Med. 25,
32503269 (2006).
31. Chuang-Stein, C., Anderson, K., Gallo, P. & Collins, S.
Sample size reestimation: a review and
recommendations. Drug Inf. J. 40, 475484 (2006).
32. Golub, H. L. The need for more efficient clinical trials.
Stat. Med. 25, 32313235 (2006).
33. Tsiatis, A. A. & Mehta, C. On the inefficiency of the
adaptive design for monitoring clinical trials.
Biometrika 90, 367378 (2003).
Acknowledgements
A special acknowledgment to W. Dere (Amgen), S. Cummings
(UCSF), A. Lee (Pfizer) and E. Berndt (MIT-CBI) for signifi-
cant contributions to discussions leading to this manuscript.
Also, many thanks to the McKinsey Trial Design Team for their
support (M. E., E. F., N. S. and T. T.).
DATABASES
OMiM:
http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=OMiM
MuckleWells syndrome | type 2 diabetes
SUppLEMEnTARy inFoRMATion
see online article: s1 (box) | s2 (box) | s3 (figure)
ALL Links Are Active in the onLine PDf
PersPecti ves
Active
Final analysis
LPFV
Interim analysis
Sample size re-estimation
Enrollment
Control
Power
n
= 0.375
= 90%
= 1.0
= 150
Power
n
= 0.375
= 90%
= 1.4
= 295
Learning
Figure 3 | re-estimating sample size while maintaining statistical power. The figure illustrates
a hypothetical example of a study in which sample size re-estimation due to uncertainty about the
standard deviation led to an increase in sample size to ensure 90% power was maintained. At the
beginning of the trial, the planned sample size was estimated at 150 patients based on a standard
deviation of 1.0. At the interim analysis, the actual standard deviation was 1.4. even though the effect
size () was as originally predicted, an increase in sample size to 295 patients would be required to
maintain 90% power. Without the sample size re-estimation, the power at the final analysis would only
be 64% and there would be much greater risk of a failed trial. LPFv, last patient first visit.
accrual rate or good longitudinal forecasting
models; efficient design and implementa-
tion software and fast computing platforms;
an infrastructure that facilitates rapid
communication across trial sites to the
central unblinded analysis centre and rapid
communication of dose assignments to trial
sites; and a flexible drug-supply process.
appropriate models, which reliably charac-
terize the longitudinal behaviour of clinical
endpoints, or the relationship between
biomarkers and endpoints, are also crucial
to the success of the modern clinical devel-
opment paradigm discussed here. Because
model assumptions often need to be
checked and at times revised after data
have been observed, an intriguing possibility
is to use adaptive modelling approaches.
This is a topic for further research, and is
beyond the scope for this paper.
Maximizing the use of all potential prior
information requires greater collaboration
across functional silos in organizations
to avoid compartmentalization of data.
In practice, the inclusion of a broader sample
of datasets can be difficult because of a lack of
common data standards. These problems are
compounded by competitive hurdles to sharing
what is considered proprietary information
about novel therapies without PoC, which
inhibits the exchange of data. overcoming
internal resistance and aversion to change also
represents a major hurdle for incorporating
the prospective use of novel trial designs and
methodologies, and modelling and simula-
tion, into clinical development programmes.
a key challenge for the implementation
of tools and techniques which advance the
quality, timeliness and efficiency of drug
development is the ability to work across dis-
ciplines and amongst stakeholders to under-
stand how and when to apply these solutions.
To address this challenge, we make the fol-
lowing recommendations. First, a common
vocabulary and a common understanding of
the value of modern trial designs to all stake-
holders needs to be defined and disseminated.
Second, at the same time, guidelines and case
studies for assessing situations in which tools
should be applied, as well as for those scenar-
ios when they should not be utilized, should
be developed and disseminated. Third, there
is a need to create a methodology for dia-
logue with regulatory authorities to facilitate
discussion of clinical strategies which utilize
these tools and address potential constraints
and issues. Finally, it will be crucial to iden-
tify specific solutions to address all mindset
obstacles and factual objections that inhibit
the adoption of modern tools and adaptive
study designs.
John Orloff, Jose Pinheiro, Michael Branson,
Paul Gallo and Donald Stanski are at Novartis
Pharmaceuticals Corporation, One Health Plaza,
East Hanover, New Jersey 07936, USA.
Frank L. Douglas is at the Austen BioInnovation
Institute, 1 South Main Street, Suite 401,
Akron, Ohio 44308, USA.
Susan Levinson

is

at The Strategic Choice LLC, 6 East
Cove Lane, Morristown, New Jersey 07960, USA.
Pravin Chaturvedi is at Napo Pharmaceuticals,
250 East Grand Avenue, Suite 70,
South San Francisco, California 94080, USA.
Ene Ette

is at Anoixis Corporation, 214 North Main
Street, Suite 104, Natick, Massachusetts 01760, USA.
Gigi Hirsch, Nitin Patel,

Sameer Sabir, Stacy Springs
and

Howard Golub

are

at the Center for Biomedical
Innovation, Massachusetts Institute of Technology,
77 Massachusetts Avenue (E19611), Cambridge,
Cyrus Mehta is at Cytel Statistical Software and
Services, 675 Massachusetts Avenue, Cambridge,
Matthias R. Evers, Edd Fleming, Navjot Singh and
Tony Tramontin are

at McKinsey & Company, Inc.,
Am Sandtorkai 77, 20457 Hamburg, Germany.
Correspondence to J. O.
email: john.orloff@novartis.com
doi:10.1038/nrd3025
Published online 9 October 2009
1. Booth, B & Zemmel, R. Prospects for productivity.
2. Gilbert, J., Henske, P. & Singh, A. Rebuilding big
pharmas business model. In Vivo 21, 14 (2003).
711716 (2004).
4. Budget US Govt., App., FY 19932003.
5. Parexels Pharmaceutical R&D Statistical Sourcebook
2002/2003 (Parexel International, Waltham, USA,
2003).
6. Adams, C. P. & Brantner, V. V. Spending on new
drug development. Health Econ. 26 Feb 2009
(doi: 10.1002/hec.1454).
7. DiMasi, J. A., Hansen, R. W. & Grabowski, H. G.
The price of innovation: new estimates of drug
development costs. J. Health Econ. 22, 151185
(2003).
8. Adams, C. & Brantner, V. V. Estimating the cost of
new drug development: is it really $802 million?
Health Aff. 2, 420428 (2006).
9. Pharmaceutical R&D Factbook 2007 (CMR
International, London, UK, 2007).
10. KMR General Metrics Study (KMR Group,
Chicago, USA, 2007).
11. Sheiner, L. B. & Steimer, J.-L. Pharmacokinetic/
pharmacodynamic modeling in drug development.
Ann. Rev. Pharmacol. Toxicol. 40, 6795 (2000).
12. Lalonde, R. L. et al. Model-based drug development.
13. Breimer, D. D. & Danhof, M. Relevance of the
application of pharmacokineticpharmacodynamic
modeling concepts in drug development. The Wooden
Shoe paradigm. Clin. Pharmacokinet. 32, 259267
(1997).
14. Danhof, M., Alvan, G., Dahl, S. G., Kuhlmann, J. &
Paintaud, G. Mechanism-based pharmacokinetic/
pharmacodynamic modeling a new classification of
biomarkers. Pharm. Res. 22, 14321437 (2005).
15. Holford, N. H. G., Kimko, H. C., Monteleone, J. P. R.
& Peck, C. C. Simulation of clinical trials. Annu. Rev.
Pharmacol. Toxicol. 40, 209234 (2000).
16. Miller, R. et al. How modeling and simulation have
enhanced decision making in new drug development.
J. Pharmacokinet. Pharmacodyn. 32, 185197
(2005).
17. Berry, D. A. Bayesian statistics. Med. Decis. Making
26, 429430 (2006).
18. Mller, P., Berry, D. A., Grieve, A. P., Smith, M. &
Krams, M. Simulation-based sequential Bayesian
design. J. Stat. Plan. Inference 137, 31403150
(2007).
19. Goggin, T. et al. Modeling and simulation of clinical
trials: an industrial perspective. In Simulation for
Designing of Clinical Trials (eds Kimko, H. C. & Duffull,
S. B.) 227224 (Marcel Dekker, New York, USA,
2002).
20. Reigner, B. G. et al. An evaluation of the integration
of pharmacokinetic and pharmacodynamic principles
in clinical drug development. Experience within
Hoffman La Roche. Clin. Pharmacokinet. 33,
142152 (1997).
J. Med. 360, 24162425 (2009).
22. Berry, D. A. Bayesian clinical trials. Nature Rev. Drug
Discov. 5, 2736 (2006).
23. Berry, D. A. Introduction to Bayesian methods III:
use and interpretation of Bayesian tools in design and
analysis. Clin. Trials 2, 295300 (2005).
24. Berry, D. A. Adaptive trial design. Clin. Adv. Hematol.
Oncol. 5, 522524 (2007).
25. Krams, M. et al. Adaptive designs in clinical drug
development: opportunities, challenges, and
scope reflections following PhRMAs November
2006 workshop. J. Biopharm. Stat.17, 957964
(2007).
PersPecti ves
26. Gallo, P. et al. Adaptive designs in clinical drug
development an executive summary of the PhRMA
Working Group. J. Biopharm. Stat. 16, 275283
(2006).
27. Bornkamp, B. et al. Innovative approaches for
designing and analyzing adaptive dose-ranging trials.
J. Biopharm. Stat. 17, 965995 (2007).
28. Weir, C. J., Spiegelhalter, D. J. & Grieve, A. P.
Flexible design and efficient implementation of
adaptive dose-finding studies. J. Biopharm. Stat.
17, 10331050 (2007).
29. Bretz, F., Schmidli, H., Knig, F., Racine, A. &
Maurer, W. Confirmatory seamless Phase II/III clinical
trials with hypotheses selection at interim: general
concepts. Biom. J. 48, 623634 (2006).
30. Mehta, C. R. & Patel, N. R. Adaptive, group
sequential and decision theoretic approaches
to sample size determination. Stat. Med. 25,
32503269 (2006).
31. Chuang-Stein, C., Anderson, K., Gallo, P. & Collins, S.
Sample size reestimation: a review and
recommendations. Drug Inf. J. 40, 475484 (2006).
32. Golub, H. L. The need for more efficient clinical trials.
Stat. Med. 25, 32313235 (2006).
33. Tsiatis, A. A. & Mehta, C. On the inefficiency of the
adaptive design for monitoring clinical trials.
Biometrika 90, 367378 (2003).
Acknowledgements
A special acknowledgment to W. Dere (Amgen), S. Cummings
(UCSF), A. Lee (Pfizer) and E. Berndt (MIT-CBI) for signifi-
cant contributions to discussions leading to this manuscript.
Also, many thanks to the McKinsey Trial Design Team for their
support (M. E., E. F., N. S. and T. T.).
DATABASES
OMiM:
http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=OMiM
MuckleWells syndrome | type 2 diabetes
SUppLEMEnTARy inFoRMATion
see online article: s1 (box) | s2 (box) | s3 (figure)
ALL Links Are Active in the onLine PDf
PersPecti ves
Active
Final analysis
LPFV
Interim analysis
Sample size re-estimation
Enrollment
Control
Power
n
= 0.375
= 90%
= 1.0
= 150
Power
n
= 0.375
= 90%
= 1.4
= 295
Learning
Figure 3 | re-estimating sample size while maintaining statistical power. The figure illustrates
a hypothetical example of a study in which sample size re-estimation due to uncertainty about the
standard deviation led to an increase in sample size to ensure 90% power was maintained. At the
beginning of the trial, the planned sample size was estimated at 150 patients based on a standard
deviation of 1.0. At the interim analysis, the actual standard deviation was 1.4. even though the effect
size () was as originally predicted, an increase in sample size to 295 patients would be required to
maintain 90% power. Without the sample size re-estimation, the power at the final analysis would only
be 64% and there would be much greater risk of a failed trial. LPFv, last patient first visit.
accrual rate or good longitudinal forecasting
models; efficient design and implementa-
tion software and fast computing platforms;
an infrastructure that facilitates rapid
communication across trial sites to the
central unblinded analysis centre and rapid
communication of dose assignments to trial
sites; and a flexible drug-supply process.
appropriate models, which reliably charac-
terize the longitudinal behaviour of clinical
endpoints, or the relationship between
biomarkers and endpoints, are also crucial
to the success of the modern clinical devel-
opment paradigm discussed here. Because
model assumptions often need to be
checked and at times revised after data
have been observed, an intriguing possibility
is to use adaptive modelling approaches.
This is a topic for further research, and is
beyond the scope for this paper.
Maximizing the use of all potential prior
information requires greater collaboration
across functional silos in organizations
to avoid compartmentalization of data.
In practice, the inclusion of a broader sample
of datasets can be difficult because of a lack of
common data standards. These problems are
compounded by competitive hurdles to sharing
what is considered proprietary information
about novel therapies without PoC, which
inhibits the exchange of data. overcoming
internal resistance and aversion to change also
represents a major hurdle for incorporating
the prospective use of novel trial designs and
methodologies, and modelling and simula-
tion, into clinical development programmes.
a key challenge for the implementation
of tools and techniques which advance the
quality, timeliness and efficiency of drug
development is the ability to work across dis-
ciplines and amongst stakeholders to under-
stand how and when to apply these solutions.
To address this challenge, we make the fol-
lowing recommendations. First, a common
vocabulary and a common understanding of
the value of modern trial designs to all stake-
holders needs to be defined and disseminated.
Second, at the same time, guidelines and case
studies for assessing situations in which tools
should be applied, as well as for those scenar-
ios when they should not be utilized, should
be developed and disseminated. Third, there
is a need to create a methodology for dia-
logue with regulatory authorities to facilitate
discussion of clinical strategies which utilize
these tools and address potential constraints
and issues. Finally, it will be crucial to iden-
tify specific solutions to address all mindset
obstacles and factual objections that inhibit
the adoption of modern tools and adaptive
study designs.
John Orloff, Jose Pinheiro, Michael Branson,
Paul Gallo and Donald Stanski are at Novartis
Pharmaceuticals Corporation, One Health Plaza,
East Hanover, New Jersey 07936, USA.
Frank L. Douglas is at the Austen BioInnovation
Institute, 1 South Main Street, Suite 401,
Akron, Ohio 44308, USA.
Susan Levinson

is

at The Strategic Choice LLC, 6 East
Cove Lane, Morristown, New Jersey 07960, USA.
Pravin Chaturvedi is at Napo Pharmaceuticals,
250 East Grand Avenue, Suite 70,
South San Francisco, California 94080, USA.
Ene Ette

is at Anoixis Corporation, 214 North Main
Street, Suite 104, Natick, Massachusetts 01760, USA.
Gigi Hirsch, Nitin Patel,

Sameer Sabir, Stacy Springs
and

Howard Golub

are

at the Center for Biomedical
Innovation, Massachusetts Institute of Technology,
77 Massachusetts Avenue (E19611), Cambridge,
Cyrus Mehta is at Cytel Statistical Software and
Services, 675 Massachusetts Avenue, Cambridge,
Matthias R. Evers, Edd Fleming, Navjot Singh and
Tony Tramontin are

at McKinsey & Company, Inc.,
Am Sandtorkai 77, 20457 Hamburg, Germany.
Correspondence to J. O.
email: john.orloff@novartis.com
doi:10.1038/nrd3025
Published online 9 October 2009
1. Booth, B & Zemmel, R. Prospects for productivity.
2. Gilbert, J., Henske, P. & Singh, A. Rebuilding big
pharmas business model. In Vivo 21, 14 (2003).
711716 (2004).
4. Budget US Govt., App., FY 19932003.
5. Parexels Pharmaceutical R&D Statistical Sourcebook
2002/2003 (Parexel International, Waltham, USA,
2003).
6. Adams, C. P. & Brantner, V. V. Spending on new
drug development. Health Econ. 26 Feb 2009
(doi: 10.1002/hec.1454).
7. DiMasi, J. A., Hansen, R. W. & Grabowski, H. G.
The price of innovation: new estimates of drug
development costs. J. Health Econ. 22, 151185
(2003).
8. Adams, C. & Brantner, V. V. Estimating the cost of
new drug development: is it really $802 million?
Health Aff. 2, 420428 (2006).
9. Pharmaceutical R&D Factbook 2007 (CMR
International, London, UK, 2007).
10. KMR General Metrics Study (KMR Group,
Chicago, USA, 2007).
11. Sheiner, L. B. & Steimer, J.-L. Pharmacokinetic/
pharmacodynamic modeling in drug development.
Ann. Rev. Pharmacol. Toxicol. 40, 6795 (2000).
12. Lalonde, R. L. et al. Model-based drug development.
13. Breimer, D. D. & Danhof, M. Relevance of the
application of pharmacokineticpharmacodynamic
modeling concepts in drug development. The Wooden
Shoe paradigm. Clin. Pharmacokinet. 32, 259267
(1997).
14. Danhof, M., Alvan, G., Dahl, S. G., Kuhlmann, J. &
Paintaud, G. Mechanism-based pharmacokinetic/
pharmacodynamic modeling a new classification of
biomarkers. Pharm. Res. 22, 14321437 (2005).
15. Holford, N. H. G., Kimko, H. C., Monteleone, J. P. R.
& Peck, C. C. Simulation of clinical trials. Annu. Rev.
Pharmacol. Toxicol. 40, 209234 (2000).
16. Miller, R. et al. How modeling and simulation have
enhanced decision making in new drug development.
J. Pharmacokinet. Pharmacodyn. 32, 185197
(2005).
17. Berry, D. A. Bayesian statistics. Med. Decis. Making
26, 429430 (2006).
18. Mller, P., Berry, D. A., Grieve, A. P., Smith, M. &
Krams, M. Simulation-based sequential Bayesian
design. J. Stat. Plan. Inference 137, 31403150
(2007).
19. Goggin, T. et al. Modeling and simulation of clinical
trials: an industrial perspective. In Simulation for
Designing of Clinical Trials (eds Kimko, H. C. & Duffull,
S. B.) 227224 (Marcel Dekker, New York, USA,
2002).
20. Reigner, B. G. et al. An evaluation of the integration
of pharmacokinetic and pharmacodynamic principles
in clinical drug development. Experience within
Hoffman La Roche. Clin. Pharmacokinet. 33,
142152 (1997).
J. Med. 360, 24162425 (2009).
22. Berry, D. A. Bayesian clinical trials. Nature Rev. Drug
Discov. 5, 2736 (2006).
23. Berry, D. A. Introduction to Bayesian methods III:
use and interpretation of Bayesian tools in design and
analysis. Clin. Trials 2, 295300 (2005).
24. Berry, D. A. Adaptive trial design. Clin. Adv. Hematol.
Oncol. 5, 522524 (2007).
25. Krams, M. et al. Adaptive designs in clinical drug
development: opportunities, challenges, and
scope reflections following PhRMAs November
2006 workshop. J. Biopharm. Stat.17, 957964
(2007).
PersPecti ves
Brought to you by
Access the insights, advice and commentary of scientists and
entrepreneurs building biotech sectors around the world.
Trade Secrets is an interactive platform aiming to inform a new generation of biotech innovators by
providing rst-person accounts from industry insiders, and by allowing readers to directly interact with
bloggers via the comments section. Our contributors hail from India, China, Latin American, North
America, Europe and beyond, and their backgrounds span the biotech spectrum. Recent posts have covered
business models, recent returns from VC funds, and the Chinese governments role in growing
biotech, to name a few.
Join the global dialogue on life science entrepreneurship.
Scan this code using your mobile device to bookmark the site:
http://blogs.nature.com/trade_secrets
Trade Secrets
A Nature Network blog by Bioentrepreneur
23034-06_Nature_Network_TradeSecrets.indd 1 14/06/2011 16:51
Large pharma is in trouble, but is essential
for the success of the industry. Tadataka Tachi
Yamada, President, Global Health Program,
Bill and Melinda Gates Foundation
The pharmaceutical industry is currently
facing the key challenges of declining R&D
productivity, higher barriers to commercial
success for innovative drugs and substantial
imminent losses of revenue from successful
products due to generic competition.
For example, it has been estimated that
for every US dollar of revenue lost from
established products by the largest pharma
ceutical companies as a group between
2007 and 2012, only 26 cents will be
replaced by revenue from new products
1
.
Awareness of these challenges has
catalysed and continues to drive
considerable reorganizations in the R&D
structures of large pharmaceutical com
panies. Among the goals of such reorganiza
tions has been the promotion of the type of
entrepreneurial culture and behaviour that
is considered to thrive in smaller biotech
nology companies
2
in the hope that this will
increase R&D productivity. Indeed, industry
observers have attributed some of the
present crises in the pharmaceutical industry
to the discouragement of entrepreneurial
behaviour by limitations inherent in the
unwieldy bureaucracies that can proliferate
in large pharmaceutical companies
3
.
Given that the R&D departments in large
pharmaceutical companies in theory provide
strong platforms for innovation and thus
competitive advantage, we therefore sought
to investigate three interrelated questions
about the potential for entrepreneurship in
such companies. First, to what extent is there
evidence of entrepreneurial behaviour in large
pharmaceutical companies? Second, can the
entrepreneurial behaviour characteristic of
small biotechnology firms coexist in the con
text of the largescale and latestage develop
ment activities typical in large pharmaceutical
companies? And third, are there any lessons
from the experiences of R&D leaders in such
companies that could be used to inform the
future development of corporate cultures in
the pharmaceutical industry?
In an attempt to answer these questions,
we first reviewed the relevant literature
and identified some of the characteristics
associated with entrepreneurial behaviour
4
.
It must, however, be noted that there is a
paucity of literature on entrepreneurial
behaviour in industries with a prolonged
cycle time between the application of new
science and product launch, as is the case in
the pharmaceutical industry. We chose to
focus on three principal characteristics of
entrepreneurs which we termed vision,
bias for action and winning attitude and
identified forms of behaviour corresponding
to each category for both individual
entrepreneurs and entrepreneurial
organizations. Vision refers to the ability
to see and the drive to realize the potential
value of nascent ideas and technologies.
Bias for action refers to a readiness to make
and to implement decisions and to modify
these actions as new information becomes
available. Winning attitude is the propensity
to see hurdles as manageable challenges and
to treat what others consider failure merely
as unwanted or unexpected outcomes.
Our research relied on openended
interviews with former and present leaders
of R&D departments in pharmaceutical and
biotechnology companies (see BOX 1 for the
affiliations of the interviewees and BOX 2
for the interview methods). Together, these
organizations represented greater than 35% of
the pharmaceutical industry output in 2007
(as measured in sales)
5,6
. In this article, we
synthesize the findings from these interviews
to highlight common themes and key factors
that could promote entrepreneurial behaviour
in the pharmaceutical industry, and thereby
help to enhance R&D productivity.
Theme 1: fewer shots on goal
Success isnt necessarily how many shots on
goal, but on betting on the high probability.
Its fewer shots on goal. Phil Needleman,
former Senior Executive Vice President and
Chief Scientist, Pharmacia
Approaches to drug discovery that focus
on the number of compounds, which were
established in some companies from the
late 1990s onwards, may have actually
discouraged entrepreneurial behaviour
during the discovery phase. The compound
progression model [which focuses on the
significance of attrition during each phase of
development] might have rendered a major
disservice to the biopharmaceutical industry,
noted Yamada. It took the industry away
from innovation and pushed it to volume.
The consequence of this approach was
a focus on portfolio management and on
quantity instead of quality, and an emphasis
on the production of new molecular entities
(NMEs) at each stage. As Yamada explained,
large companies were then evaluated and
rewarded by analysts for the number of
compounds in their R&D pipeline and,
OuTlOOk
The case for entrepreneurship in
R&D in the pharmaceutical industry
Frank L. Douglas, V. K. Narayanan, Lesa Mitchell and Robert E. Litan
Abstract | A lack of entrepreneurial behaviour has often been highlighted as a
contributor to the decline in the research and development (R&D) productivity
of the pharmaceutical industry. Here, we present an assessment of
entrepreneurship in the industry, based on interviews with 26 former and
current leaders of R&D departments at major pharmaceutical and biotechnology
companies. Factors are highlighted that could be important in promoting
entrepreneurial behaviour, which might serve as a catalyst for revitalizing R&D
productivity.
PeRsPectives
NATURE REVIEWS | Drug Discovery VOlUME 9 | SEPTEMBER 2010 | 683
nrd_persp_sep10.indd 683 18/8/10 14:06:26
Brought to you by
Access the insights, advice and commentary of scientists and
entrepreneurs building biotech sectors around the world.
Trade Secrets is an interactive platform aiming to inform a new generation of biotech innovators by
providing rst-person accounts from industry insiders, and by allowing readers to directly interact with
bloggers via the comments section. Our contributors hail from India, China, Latin American, North
America, Europe and beyond, and their backgrounds span the biotech spectrum. Recent posts have covered
business models, recent returns from VC funds, and the Chinese governments role in growing
biotech, to name a few.
Join the global dialogue on life science entrepreneurship.
Scan this code using your mobile device to bookmark the site:
http://blogs.nature.com/trade_secrets
Trade Secrets
A Nature Network blog by Bioentrepreneur
23034-06_Nature_Network_TradeSecrets.indd 1 14/06/2011 16:51
Large pharma is in trouble, but is essential
for the success of the industry. Tadataka Tachi
Yamada, President, Global Health Program,
Bill and Melinda Gates Foundation
The pharmaceutical industry is currently
facing the key challenges of declining R&D
productivity, higher barriers to commercial
success for innovative drugs and substantial
imminent losses of revenue from successful
products due to generic competition.
For example, it has been estimated that
for every US dollar of revenue lost from
established products by the largest pharma
ceutical companies as a group between
2007 and 2012, only 26 cents will be
replaced by revenue from new products
1
.
Awareness of these challenges has
catalysed and continues to drive
considerable reorganizations in the R&D
structures of large pharmaceutical com
panies. Among the goals of such reorganiza
tions has been the promotion of the type of
entrepreneurial culture and behaviour that
is considered to thrive in smaller biotech
nology companies
2
in the hope that this will
increase R&D productivity. Indeed, industry
observers have attributed some of the
present crises in the pharmaceutical industry
to the discouragement of entrepreneurial
behaviour by limitations inherent in the
unwieldy bureaucracies that can proliferate
in large pharmaceutical companies
3
.
Given that the R&D departments in large
pharmaceutical companies in theory provide
strong platforms for innovation and thus
competitive advantage, we therefore sought
to investigate three interrelated questions
about the potential for entrepreneurship in
such companies. First, to what extent is there
evidence of entrepreneurial behaviour in large
pharmaceutical companies? Second, can the
entrepreneurial behaviour characteristic of
small biotechnology firms coexist in the con
text of the largescale and latestage develop
ment activities typical in large pharmaceutical
companies? And third, are there any lessons
from the experiences of R&D leaders in such
companies that could be used to inform the
future development of corporate cultures in
the pharmaceutical industry?
In an attempt to answer these questions,
we first reviewed the relevant literature
and identified some of the characteristics
associated with entrepreneurial behaviour
4
.
It must, however, be noted that there is a
paucity of literature on entrepreneurial
behaviour in industries with a prolonged
cycle time between the application of new
science and product launch, as is the case in
the pharmaceutical industry. We chose to
focus on three principal characteristics of
entrepreneurs which we termed vision,
bias for action and winning attitude and
identified forms of behaviour corresponding
to each category for both individual
entrepreneurs and entrepreneurial
organizations. Vision refers to the ability
to see and the drive to realize the potential
value of nascent ideas and technologies.
Bias for action refers to a readiness to make
and to implement decisions and to modify
these actions as new information becomes
available. Winning attitude is the propensity
to see hurdles as manageable challenges and
to treat what others consider failure merely
as unwanted or unexpected outcomes.
Our research relied on openended
interviews with former and present leaders
of R&D departments in pharmaceutical and
biotechnology companies (see BOX 1 for the
affiliations of the interviewees and BOX 2
for the interview methods). Together, these
organizations represented greater than 35% of
the pharmaceutical industry output in 2007
(as measured in sales)
5,6
. In this article, we
synthesize the findings from these interviews
to highlight common themes and key factors
that could promote entrepreneurial behaviour
in the pharmaceutical industry, and thereby
help to enhance R&D productivity.
Theme 1: fewer shots on goal
Success isnt necessarily how many shots on
goal, but on betting on the high probability.
Its fewer shots on goal. Phil Needleman,
former Senior Executive Vice President and
Chief Scientist, Pharmacia
Approaches to drug discovery that focus
on the number of compounds, which were
established in some companies from the
late 1990s onwards, may have actually
discouraged entrepreneurial behaviour
during the discovery phase. The compound
progression model [which focuses on the
significance of attrition during each phase of
development] might have rendered a major
disservice to the biopharmaceutical industry,
noted Yamada. It took the industry away
from innovation and pushed it to volume.
The consequence of this approach was
a focus on portfolio management and on
quantity instead of quality, and an emphasis
on the production of new molecular entities
(NMEs) at each stage. As Yamada explained,
large companies were then evaluated and
rewarded by analysts for the number of
compounds in their R&D pipeline and,
OuTlOOk
The case for entrepreneurship in
R&D in the pharmaceutical industry
Frank L. Douglas, V. K. Narayanan, Lesa Mitchell and Robert E. Litan
Abstract | A lack of entrepreneurial behaviour has often been highlighted as a
contributor to the decline in the research and development (R&D) productivity
of the pharmaceutical industry. Here, we present an assessment of
entrepreneurship in the industry, based on interviews with 26 former and
current leaders of R&D departments at major pharmaceutical and biotechnology
companies. Factors are highlighted that could be important in promoting
entrepreneurial behaviour, which might serve as a catalyst for revitalizing R&D
productivity.
PeRsPectives
First published in Nature Reviews Drug Discovery 9, 683-689 (September 2010); doi:10.1038/nrd3230
Theme 2: smaller is better
I think the key is creating something which
is close to a biotech company within big
pharma. Garry Neil, Corporate Vice President,
Corporate Office of Science and Technology
(COSAT), Group President, Johnson & Johnson
pharmaceutical R&D
Several heads of research departments
expressed the view that small is better.
Frank Douglas and Tachi Yamada each tried
to reduce size and complexity in their com
panies. In 1998, Douglas introduced the drug
innovation and approval (DI&A) organiza
tion at Hoechst Marion Roussel (now part
of SanofiAventis) in which each of the three
discovery sites (one each in New Jersey, USA,
Frankfurt, Germany, and Paris, France) was
responsible for no more than two to three
therapeutic areas through to Phase IIa, and
a global development centre coordinated the
latestage development and regulatory sub
mission activities. A product development
committee determined which compounds
would be further developed beyond Phase IIa
by the global development centre.
Yamada introduced the Centres of
Excellence for Drug Discovery (CEDDs)
at GlaxoSmithKline. Each CEDD was
responsible for one or two therapeutic areas
through to Phase IIa trials. There was more
autonomy in the CEDD, as compared to
the DI&A, in that the only item that was
centrally controlled was headcount. Recently,
GlaxoSmithKline evolved the CEDDs into
smaller subgroups, as they thought the
CEDDs had become too bureaucratic.
There seems to be an inverse correlation
between the size of an organization and its
potential to develop trust among its constitu
ents and facilitate the rapid exchange of data
for the generation and testing of hypotheses.
Joshua Boger and Vicki Sato described the
focus of Vertex Pharmaceuticals as building
a culture in which people are rewarded for
making quality data rapidly available and
ready for integration. I think they [large
pharmaceutical companies] have difficulty
sharing information quickly holding onto
information can be a form of power it is
part of a behaviour pattern that contributes
to things taking longer, costing more and
creating redundancy, said Sato.
Confidence in the quality of data pre
sented allowed departments to circumvent
the wasteful practice of rechecking data
because of a lack of trust in the expertise or
in the care of those who initially generated
the data. The accessibility of senior managers
and their involvement in project status
discussions is essential in reinforcing these
principles of rapid exchange of quality data
and testing of hypotheses, all of which are
hampered as organizations increase in size.
Boger and Sato also noted that as Vertex
increased in size, much more time was
required to build these cultural aspects. Sato
added: I went to every project team meeting
and before I joined Vertex, Josh went to
every team meeting. It allowed us as senior
executives to lead by example and build the
culture. You can do that when you are small.
However, with 200 to 250 researchers
per therapeutic area, the DI&A and CEDD
groups were still too large in the view of
Goodman, who feels that the optimum size
of a discovery group is one in which all
members could assemble in a modestsized
room and conduct a discussion without
recourse to a microphone. Corr, who led
groups at Searle, ParkeDavis and Pfizer,
thinks that the optimal number is closer to 20
to 40 individuals operating as an autonomous
group. Corr discovered that very small teams
at ParkeDavis allowed for the rapid sharing
of information and resource allocation,
as well as the efficient termination of projects
that would eventually fail. Yamada similarly
talked about small groups being agile and
interconnected to the other person.
A recent experiment at Pfizer, known
as the Biotherapeutics and Bioinnovation
Center, which was led by Goodman and
has since been dissolved, was focused on
managing four to five independent small
biotechnology companies in a network.
In this arrangement, best practices could
be shared and resources leveraged, but each
unit remained autonomous and was to
some extent free to advance its own culture.
The head of each small company was there
fore endowed with a sense of ownership and
could more or less operate as an entrepreneur.
According to Goodman, Pfizer wanted to see
whether such skunkworks strategies, similar
to those used in the information technology
and electrical engineering fields, could be
harnessed to change the way compounds
were typically developed. Neil also supports
this view of the potential of small companies
to produce changes within their respective
larger corporate structures. He characterizes
the culture at Johnson & Johnson as
a hybrid between biotech and large pharma
and attributes its success to the strategy of
maintaining the cultures of newly acquired
companies and adopting their best practices.
Contributing to the apparent consensus on
behalf of the entrepreneurial potential inher
ent in smaller teams, Fishman offered his
perspective on the importance of individuals
in innovation: To me, innovation is generally
an individual effort. There is a person who
sees something different, sees something
new, has a clarity of vision, and the courage
to pursue it. That individual expands to work
in an entrepreneurial team to bring the inno
vation to fruition. Alan Smith expressed a
similar view, noting that in his opinion it is
indeed the curiosity of the individual scien
tist that is most often the driver for discovery
efforts. Robert Armstrong, in commenting
on CHORUS a unit of 32 lilly employees
who comprise an autonomous entity he heads
that conducts lean development from the
candidate selection stage to the proofof
concept of selected lilly compounds said:
not all of them need to be serial entrepre
neurs, but there are individuals who perceive
radically different ways to develop new
molecules and want to enable these concepts
(see also the section on middle managers).
Although many of those interviewed
spoke on behalf of smaller and more entre
preneurial discovery units, Burt Adelman
pointed out that such scale does not in itself
necessarily engender the sense of urgency
crucial to successful product innovation.
He argued that entrepreneurial behaviour
diminished even in biotechnology
companies when the focus was primarily
on publications rather than on products.
In this context, he poses the difficult question
of why scientists who have not been associ
ated even with a failed drug after 6 or 7 years
should still be employed by that company.
George Milne conceded the previously
cited benefits associated with small size, but
also presented an argument on behalf of larger
corporate structures in which smaller units
benefit from the experiences and tacit aggre
gate knowledge of the larger organization.
Milne considers that the biopharmaceutical
industry must learn to scale small entrepre
neurial units without introducing excessive
bureaucracy. Such units thus become empow
ered to unleash random and purposeful
action. Milne also noted that one of the scale
based advantages that a large, entrepreneurial
organization has over a small one is that you
actually get to learn from experiences.
A comment from Jeff leiden provides an
eloquent if blunt summary to a consideration
of the benefits associated with the deploy
ment of smaller teams and the instinctive,
curiositydriven culture they foster: So, it
became clear to me that the way to organize
R&D, and frankly I think the whole busi
ness, is in much smaller entrepreneurial
units where people feel both responsible and
accountable for producing or their unit wont
be renewed. They know that their careers
depend directly on their productivity.
PersPecti ves
more recently, for the number in the later
phases of clinical development in particular.
Similarly, research organizations in large
pharmaceutical companies were rewarded
for the number of NMEs produced as
clinical candidates each year. As Corey
Goodman commented, scientists are getting
their bonuses based on trying to make the
numbers. However, with the exception of
a sharp increase between 1996 and 1998,
the number of NMEs approved annually
(FDA) has not changed over the past five
decades, even though the costs of achieving
approval have increased manyfold during
this period
7
.
Interestingly, an assessment of the value
of small biotechnology companies was driven
by the innovation and science in the one or
two products moving through their pipe
lines. As an example, Tom Glenn recounted
the birth of DNase at Genentech, which was
developed during a biweekly staff meeting
in which research ideas were discussed:
He [Steve Shak, M.D.] brought in two test
tubes. In one test tube, he had this gunky
sputum. In the other test tube, he had this
clear liquid. Ill never forget this. There were
six of us sitting there watching. He mixed
the contents of the two test tubes together,
and they cleared. He said, Gentlemen,
thats DNase. Unbelievable. On the spot we
decided to move forward and finish our
discovery and development program in
DNase, which is now a product marketed
by Genentech.
The insidious consequence of the
focus on quantity in large pharmaceutical
companies has been an emphasis on the
commercially driven evaluation of a portfolio
of compounds, instead of the scientific
merits of each compound. Mark Fishman
similarly expressed scepticism regarding
the deep involvement of commercial
departments before the availability of
clinical data from Phase II trials, when
he stated: Often, they would ask if you
are aligned with the business franchises.
I dont want to be because they are
looking at today, and I want to be there
five to ten years from now. So, if we are
aligned with the current business
franchise, we are dead in the water.
Despite such scepticism, however,
there are those scientists, such as Andreas
Busch, who although not ignoring
compound quality, remain practitioners
of the compound progression approach,
illustrating that innovation can coincide
with less entrepreneurialfocused
approaches. Peter Corr supports Buschs
view, especially when there are more shots
on goal. However, Goodman thinks
that in many of these companies, too many
drugs are being advanced into the clinic.
And Needleman was of the opinion that
success isnt necessarily how many shots
on goal, but on betting on the high
probability. Its fewer shots on goal.

Box 1 | Interviewees
Our research involved open-ended interviews with former and present leaders of research
and development (R&D) departments in large pharmaceutical companies and biotechnology
companies. The sample of interviewees included the following individuals:
Burt Adelman M.D., President of Research and Development at Eleven Biotherapeutics
(2009present), former Executive Vice President of Portfolio Strategy at Biogen Idec
(20032006).
Robert Armstrong Ph.D., Vice President of Global External Research and Development at
Lilly Research Laboratories (2006present).
Lee Babiss Ph.D., Executive Vice President, Global Laboratory Services at Pharmaceutical
Product Development (2010present), former Global Head of Pharmaceutical Research at
Hoffman-LaRoche (20072009).
Joshua Boger Ph.D., founder and former President and CEO of Vertex Pharmaceuticals
(19922009).
Andreas Busch Ph.D., Head of Global Drug Discovery at Bayer Schering Pharma (2007present).
Peter Corr Ph.D., former Senior Vice President of Science and Technology at Pfizer (20022006).
Frank Douglas Ph.D., M.D., former Executive Vice President at Aventis (19952004).
Mark Fishman M.D., President of Novartis Institutes of Biomedical Research (2002present).
Tom Glenn Ph.D., former Vice President of Pharmaceutical Sciences at Genentech (19881990).
Corey Goodman Ph.D., former Head of Biotherapeutics and Bioinnovation Center at Pfizer
(20072009).
Bernd Kirschbaum Ph.D., Executive Vice President of Global Research and Development at
Merck Serono (2008present).
Jeff Leiden Ph.D., former President and Chief Operating Officer, Pharmaceutical Products
Group at Abbott (20012006).
George Milne Ph.D., former Executive Vice President of Global Research and Development at
Pfizer (20002002).
Phil Needleman Ph.D., former Senior Executive Vice President and Chief Scientist at Pharmacia
(20002003).
Garry Neil M.D., Corporate Vice President, Corporate Office of Science and Technology at
Johnson & Johnson (2007present).
John Patterson Ph.D., former Executive Director of Development at AstraZeneca (20052009).
Steven Paul M.D., former Executive Vice President of Science and Technology and President of
Lilly Research Laboratories at Eli Lilly (20032010).
Joerg Reinhardt Ph.D., Chief Operating Officer at Novartis (20002009).
Peter Ringrose Ph.D., former President of the Bristol-Myers Squibb Pharmaceutical Research
Institute at Bristol-Myers Squibb (19972003).
David Rosen D.V.M., Executive Director and Head of Out Licensing, Worldwide Business
Development at Pfizer (20072009).
Leon Rosenberg M.D., Professor in the Department of Molecular Biology at Princeton University,
New Jersey, USA (1997present), former Senior Vice President of Scientific Affairs at
Bristol-Myers Squibb (19911997).
Robert Ruffolo Ph.D., former President of Research and Development at Wyeth (20022008).
Vicki Sato Ph.D., Professor of Management Practice and Professor of the Practice in the
Department of Molecular and Cell Biology at Harvard University, Boston, Massachusetts, USA
(2005present), former President of Vertex Pharmaceuticals (20002005).
Ben Shapiro M.D., former Executive Vice President, Worldwide Licensing and External Research
at Merck (19962003).
Alan Smith Ph.D., Senior Vice President of Research and Chief Scientific Officer at Genzyme
(1996present).
Gus Watanabe M.D. (now deceased), former Executive Vice President of Science and Technology
at Eli Lilly (19942003).
Tachi Yamada M.D., President, Global Health Program at the Bill and Melinda Gates Foundation
(2006present), former Chairman of R&D, GlaxoSmithKline (20012006).
PersPecti ves
684 | SEPTEMBER 2010 | VOlUME 9 www.nature.com/reviews/drugdisc
Theme 2: smaller is better
I think the key is creating something which
is close to a biotech company within big
pharma. Garry Neil, Corporate Vice President,
Corporate Office of Science and Technology
(COSAT), Group President, Johnson & Johnson
pharmaceutical R&D
Several heads of research departments
expressed the view that small is better.
Frank Douglas and Tachi Yamada each tried
to reduce size and complexity in their com
panies. In 1998, Douglas introduced the drug
innovation and approval (DI&A) organiza
tion at Hoechst Marion Roussel (now part
of SanofiAventis) in which each of the three
discovery sites (one each in New Jersey, USA,
Frankfurt, Germany, and Paris, France) was
responsible for no more than two to three
therapeutic areas through to Phase IIa, and
a global development centre coordinated the
latestage development and regulatory sub
mission activities. A product development
committee determined which compounds
would be further developed beyond Phase IIa
by the global development centre.
Yamada introduced the Centres of
Excellence for Drug Discovery (CEDDs)
at GlaxoSmithKline. Each CEDD was
responsible for one or two therapeutic areas
through to Phase IIa trials. There was more
autonomy in the CEDD, as compared to
the DI&A, in that the only item that was
centrally controlled was headcount. Recently,
GlaxoSmithKline evolved the CEDDs into
smaller subgroups, as they thought the
CEDDs had become too bureaucratic.
There seems to be an inverse correlation
between the size of an organization and its
potential to develop trust among its constitu
ents and facilitate the rapid exchange of data
for the generation and testing of hypotheses.
Joshua Boger and Vicki Sato described the
focus of Vertex Pharmaceuticals as building
a culture in which people are rewarded for
making quality data rapidly available and
ready for integration. I think they [large
pharmaceutical companies] have difficulty
sharing information quickly holding onto
information can be a form of power it is
part of a behaviour pattern that contributes
to things taking longer, costing more and
creating redundancy, said Sato.
Confidence in the quality of data pre
sented allowed departments to circumvent
the wasteful practice of rechecking data
because of a lack of trust in the expertise or
in the care of those who initially generated
the data. The accessibility of senior managers
and their involvement in project status
discussions is essential in reinforcing these
principles of rapid exchange of quality data
and testing of hypotheses, all of which are
hampered as organizations increase in size.
Boger and Sato also noted that as Vertex
increased in size, much more time was
required to build these cultural aspects. Sato
added: I went to every project team meeting
and before I joined Vertex, Josh went to
every team meeting. It allowed us as senior
executives to lead by example and build the
culture. You can do that when you are small.
However, with 200 to 250 researchers
per therapeutic area, the DI&A and CEDD
groups were still too large in the view of
Goodman, who feels that the optimum size
of a discovery group is one in which all
members could assemble in a modestsized
room and conduct a discussion without
recourse to a microphone. Corr, who led
groups at Searle, ParkeDavis and Pfizer,
thinks that the optimal number is closer to 20
to 40 individuals operating as an autonomous
group. Corr discovered that very small teams
at ParkeDavis allowed for the rapid sharing
of information and resource allocation,
as well as the efficient termination of projects
that would eventually fail. Yamada similarly
talked about small groups being agile and
interconnected to the other person.
A recent experiment at Pfizer, known
as the Biotherapeutics and Bioinnovation
Center, which was led by Goodman and
has since been dissolved, was focused on
managing four to five independent small
biotechnology companies in a network.
In this arrangement, best practices could
be shared and resources leveraged, but each
unit remained autonomous and was to
some extent free to advance its own culture.
The head of each small company was there
fore endowed with a sense of ownership and
could more or less operate as an entrepreneur.
According to Goodman, Pfizer wanted to see
whether such skunkworks strategies, similar
to those used in the information technology
and electrical engineering fields, could be
harnessed to change the way compounds
were typically developed. Neil also supports
this view of the potential of small companies
to produce changes within their respective
larger corporate structures. He characterizes
the culture at Johnson & Johnson as
a hybrid between biotech and large pharma
and attributes its success to the strategy of
maintaining the cultures of newly acquired
companies and adopting their best practices.
Contributing to the apparent consensus on
behalf of the entrepreneurial potential inher
ent in smaller teams, Fishman offered his
perspective on the importance of individuals
in innovation: To me, innovation is generally
an individual effort. There is a person who
sees something different, sees something
new, has a clarity of vision, and the courage
to pursue it. That individual expands to work
in an entrepreneurial team to bring the inno
vation to fruition. Alan Smith expressed a
similar view, noting that in his opinion it is
indeed the curiosity of the individual scien
tist that is most often the driver for discovery
efforts. Robert Armstrong, in commenting
on CHORUS a unit of 32 lilly employees
who comprise an autonomous entity he heads
that conducts lean development from the
candidate selection stage to the proofof
concept of selected lilly compounds said:
not all of them need to be serial entrepre
neurs, but there are individuals who perceive
radically different ways to develop new
molecules and want to enable these concepts
(see also the section on middle managers).
Although many of those interviewed
spoke on behalf of smaller and more entre
preneurial discovery units, Burt Adelman
pointed out that such scale does not in itself
necessarily engender the sense of urgency
crucial to successful product innovation.
He argued that entrepreneurial behaviour
diminished even in biotechnology
companies when the focus was primarily
on publications rather than on products.
In this context, he poses the difficult question
of why scientists who have not been associ
ated even with a failed drug after 6 or 7 years
should still be employed by that company.
George Milne conceded the previously
cited benefits associated with small size, but
also presented an argument on behalf of larger
corporate structures in which smaller units
benefit from the experiences and tacit aggre
gate knowledge of the larger organization.
Milne considers that the biopharmaceutical
industry must learn to scale small entrepre
neurial units without introducing excessive
bureaucracy. Such units thus become empow
ered to unleash random and purposeful
action. Milne also noted that one of the scale
based advantages that a large, entrepreneurial
organization has over a small one is that you
actually get to learn from experiences.
A comment from Jeff leiden provides an
eloquent if blunt summary to a consideration
of the benefits associated with the deploy
ment of smaller teams and the instinctive,
curiositydriven culture they foster: So, it
became clear to me that the way to organize
R&D, and frankly I think the whole busi
ness, is in much smaller entrepreneurial
units where people feel both responsible and
accountable for producing or their unit wont
be renewed. They know that their careers
depend directly on their productivity.
PersPecti ves
more recently, for the number in the later
phases of clinical development in particular.
Similarly, research organizations in large
pharmaceutical companies were rewarded
for the number of NMEs produced as
clinical candidates each year. As Corey
Goodman commented, scientists are getting
their bonuses based on trying to make the
numbers. However, with the exception of
a sharp increase between 1996 and 1998,
the number of NMEs approved annually
(FDA) has not changed over the past five
decades, even though the costs of achieving
approval have increased manyfold during
this period
7
.
Interestingly, an assessment of the value
of small biotechnology companies was driven
by the innovation and science in the one or
two products moving through their pipe
lines. As an example, Tom Glenn recounted
the birth of DNase at Genentech, which was
developed during a biweekly staff meeting
in which research ideas were discussed:
He [Steve Shak, M.D.] brought in two test
tubes. In one test tube, he had this gunky
sputum. In the other test tube, he had this
clear liquid. Ill never forget this. There were
six of us sitting there watching. He mixed
the contents of the two test tubes together,
and they cleared. He said, Gentlemen,
thats DNase. Unbelievable. On the spot we
decided to move forward and finish our
discovery and development program in
DNase, which is now a product marketed
by Genentech.
The insidious consequence of the
focus on quantity in large pharmaceutical
companies has been an emphasis on the
commercially driven evaluation of a portfolio
of compounds, instead of the scientific
merits of each compound. Mark Fishman
similarly expressed scepticism regarding
the deep involvement of commercial
departments before the availability of
clinical data from Phase II trials, when
he stated: Often, they would ask if you
are aligned with the business franchises.
I dont want to be because they are
looking at today, and I want to be there
five to ten years from now. So, if we are
aligned with the current business
franchise, we are dead in the water.
Despite such scepticism, however,
there are those scientists, such as Andreas
Busch, who although not ignoring
compound quality, remain practitioners
of the compound progression approach,
illustrating that innovation can coincide
with less entrepreneurialfocused
approaches. Peter Corr supports Buschs
view, especially when there are more shots
on goal. However, Goodman thinks
that in many of these companies, too many
drugs are being advanced into the clinic.
And Needleman was of the opinion that
success isnt necessarily how many shots
on goal, but on betting on the high
probability. Its fewer shots on goal.

Box 1 | Interviewees
Our research involved open-ended interviews with former and present leaders of research
and development (R&D) departments in large pharmaceutical companies and biotechnology
companies. The sample of interviewees included the following individuals:
Burt Adelman M.D., President of Research and Development at Eleven Biotherapeutics
(2009present), former Executive Vice President of Portfolio Strategy at Biogen Idec
(20032006).
Robert Armstrong Ph.D., Vice President of Global External Research and Development at
Lilly Research Laboratories (2006present).
Lee Babiss Ph.D., Executive Vice President, Global Laboratory Services at Pharmaceutical
Product Development (2010present), former Global Head of Pharmaceutical Research at
Hoffman-LaRoche (20072009).
Joshua Boger Ph.D., founder and former President and CEO of Vertex Pharmaceuticals
(19922009).
Andreas Busch Ph.D., Head of Global Drug Discovery at Bayer Schering Pharma (2007present).
Peter Corr Ph.D., former Senior Vice President of Science and Technology at Pfizer (20022006).
Frank Douglas Ph.D., M.D., former Executive Vice President at Aventis (19952004).
Mark Fishman M.D., President of Novartis Institutes of Biomedical Research (2002present).
Tom Glenn Ph.D., former Vice President of Pharmaceutical Sciences at Genentech (19881990).
Corey Goodman Ph.D., former Head of Biotherapeutics and Bioinnovation Center at Pfizer
(20072009).
Bernd Kirschbaum Ph.D., Executive Vice President of Global Research and Development at
Merck Serono (2008present).
Jeff Leiden Ph.D., former President and Chief Operating Officer, Pharmaceutical Products
Group at Abbott (20012006).
George Milne Ph.D., former Executive Vice President of Global Research and Development at
Pfizer (20002002).
Phil Needleman Ph.D., former Senior Executive Vice President and Chief Scientist at Pharmacia
(20002003).
Garry Neil M.D., Corporate Vice President, Corporate Office of Science and Technology at
Johnson & Johnson (2007present).
John Patterson Ph.D., former Executive Director of Development at AstraZeneca (20052009).
Steven Paul M.D., former Executive Vice President of Science and Technology and President of
Lilly Research Laboratories at Eli Lilly (20032010).
Joerg Reinhardt Ph.D., Chief Operating Officer at Novartis (20002009).
Peter Ringrose Ph.D., former President of the Bristol-Myers Squibb Pharmaceutical Research
Institute at Bristol-Myers Squibb (19972003).
David Rosen D.V.M., Executive Director and Head of Out Licensing, Worldwide Business
Development at Pfizer (20072009).
Leon Rosenberg M.D., Professor in the Department of Molecular Biology at Princeton University,
New Jersey, USA (1997present), former Senior Vice President of Scientific Affairs at
Bristol-Myers Squibb (19911997).
Robert Ruffolo Ph.D., former President of Research and Development at Wyeth (20022008).
Vicki Sato Ph.D., Professor of Management Practice and Professor of the Practice in the
Department of Molecular and Cell Biology at Harvard University, Boston, Massachusetts, USA
(2005present), former President of Vertex Pharmaceuticals (20002005).
Ben Shapiro M.D., former Executive Vice President, Worldwide Licensing and External Research
at Merck (19962003).
Alan Smith Ph.D., Senior Vice President of Research and Chief Scientific Officer at Genzyme
(1996present).
Gus Watanabe M.D. (now deceased), former Executive Vice President of Science and Technology
at Eli Lilly (19942003).
Tachi Yamada M.D., President, Global Health Program at the Bill and Melinda Gates Foundation
(2006present), former Chairman of R&D, GlaxoSmithKline (20012006).
PersPecti ves
also introduced the proofofconcept unit
CHORUS (see above), and in an extension of
this concept, lilly launched a joint venture
with Jubilant Organosys in Bangalore, India,
to develop molecules between preclinical
and Phase II testing in May 2009. Through
this collaboration, lilly also benefits from
the potential opportunity to have access to
successful proofofconcept compounds
from sources other than their own research
laboratories. Armstrong said that lilly has
expanded this concept to several companies
in India and China, who work on selected
compounds that lilly has the right to buy
back when predetermined milestones
are achieved. These activities dismantle
vestiges of the not invented here
syndrome that can mean certain death to
entrepreneurial behaviour, he noted.
Paul of Eli lilly summarizes this more
broadly: So, one element of the trans
formation from a corporate perspective
structurally is to go from this FIPCO (Fully
Integrated Pharmaceutical Company)
structure to a FIPNET, Fully Integrated
Pharmaceutical Network. Weve defined
three levels of this FIPNET. level one is
more traditional outsourcing. level two
leverages partnerships with universities and
small as well as larger companies. These
valuesharing partnerships provide both risk
sharing and cost sharing. Finally, a level
three FIPNET or partial ownership model
allows small companies to pursue projects
within the context of their unique cultures.
Our interviews also suggest that discovery
stage collaborations are perceived to be
effective when the collaborators share a
common sense of curiosity and focus on
science. The importance of curiosity as a
driver for the selection of technology, bio
logical mechanisms and collaborations was
stressed as a key component of the culture at
Genzyme by Smith, who added that curiosity
refers to a desire to learn and benefit from
knowledge that others in that university in
India or in France might deem interesting.
We determined that among the tasks
that need to be performed at the middle
management level are the maintenance of
organizational knowledge, the nurturing
of mavericks and true believers, and the
provision of support for the drug discovery
teams. Essential to these roles is the need
to remain close to science.
One of the distinguishing characteristics
of the biotechnology researchers we inter
viewed was their closeness to science and
their desire to retain this proximity. In large
pharmaceutical companies, however, exces
sive concern with milestones often detracts
from the ability of research teams to remain
close to science. Yet to the extent that the
rewards system associated with middle
management is based on the attainment
of milestones and not on the nurturing of
science, bureaucracy imposes barriers that
are detrimental to drug innovation. Indeed,
evolving a more sophisticated and nuanced
reward system for middle management may
be among the most important organizational
priorities for large pharmaceutical compa
nies to stimulate productivity in the crucial
discovery stage.
Theme 5: CEOHead of R&D interaction
You have to really be able to communicate
with your CEO if youre going to do your
job optimally in R&D. Gus Watanabe,
former Executive Vice President of Science
and Technology, Eli lilly
Several of the interviewees, including leon
Rosenberg, Watanabe, Goodman, Milne and
Glenn, stressed the importance of a close
alignment between the CEO and the head of
R&D in building and maintaining an optimal
culture in discovery. Boger described building
an innovative and productive organization as
more of a social experiment than a technical
one. He and Sato described in nearly identical
terms their tireless and consistent efforts to
build, nurture and maintain elements such
as interdisciplinary teamwork, rapid access
to information and a sense of ownership
in their organization. Rosenberg described
the weekly meetings that he used to have
with the CEO: He wasnt a scientist, but he
appreciated science, and he liked scientists.
He and I would converse regularly after
he would read his weekly copy of The New
England Journal of Medicine. This was a great
joy of mine because it kept me very close to
him. It allowed me to put science high in
the order of his thoughts. Smith described
similar efforts and alignment with Genzyme
CEO Henri Termeer in maintaining their
organizational culture. Busch and Bernd
Kirschbaum gave examples of the consider
able challenges and efforts needed to build a
common culture in a newly merged organi
zation and its impact on innovation, and
the important role that alignment with their
CEOs had in their efforts to build the desired
culture in discovery. Douglas also com
mented on the importance of the close align
ment between him and Richard Markham
when they took on the task of building
Hoechst Marion Roussel, the result of merg
ing Hoechst Pharmaceutical, Marion Merrell
Dow and Roussel Uclaf, in their respective
roles as Head of R&D and CEO.
Theme 6: technology and R&D models
My major responsibilities as given to me by
the CEO when I joined BMS were to forge a
single R&D organization out of two groups
with differing strengths in Discovery and
Development. leon Rosenberg, Professor at
Princeton University and former Senior Vice
President of Scientific Affairs, BristolMyers
Squibb
The interviews revealed the unexpected
finding that although each R&D leader
recognized that changing science and tech
nology influenced how they organized their
research units, most had not reflected on the
impact that organizational changes might
have on entrepreneurial behaviour. The
interviewees identified four types of organi
zational structures or models that defined
the point at which commitment was made
to the fullscale clinical development of a
compound (FIG. 1).
First, they described a traditional R&D
organization, in which the transition from
research to development occurred with
the filing of an investigational new drug
application. This is termed the preclinical
proofofconcept (pPoC) model. The second
structure mentioned the human proofof
concept (hPoC) model was similar to this,
with the addition of clinical pharmacology
to improve dose selection and the search for
efficacy in humans.
The third structure, known as the
clinical proofofconcept (cPoC) model,
includes Phase I and Phase IIa trials in
research. It uses biomarkers and pharmaco
kinetic studies to select the best compound
to determine efficacy in the target patients,
with efficacy in Phase IIa trials being the
crucial decision point for full development
(FIG. 1).
Watanabe, deceased since the interviews,
introduced this concept by having M.D.s or
M.D./Ph.D.s responsible for each discovery
therapeutic unit from the concept stage to
Phase IIa trials at Eli lilly. Glenn introduced
the clinical biology unit concept in research at
Ciba Geigy in 1984 with the goal of carrying
out specific studies in patients to achieve
proof of concept. Several other companies
have adopted this model in various forms.
These include the research and early devel
opment (RED) units at Johnson & Johnson,
the research units at MerckSerono, global
drug discovery at Bayer Schering Pharma
and disease biology areas at Roche (it
should be noted that Roches new focus on
personalized healthcare by interweaving
their diagnostic and pharmaceutical divi
sions has moved them to the fourth model;
PersPecti ves
Theme 3: reward systems matter
I am firmly a believer that you get what
you reward. Josh Boger, former CEO,
Vertex Pharmaceuticals
In most industries, as companies grow, there
is pressure for each to offer uniform com
pensation, rewards and benefits. Fishman
and Goodman, as well as Peter Ringrose,
each referred to the tendency for all depart
ments to compensate in a similar fashion as
a homogenizing effect. What is curious is
that although large companies recognize the
need to implement special bonus systems
to incentivize sale representatives to act
entrepreneurially, the opposite occurs in
R&D. The lack of incentives has persisted
owing to organizational inertia, but thought
ful R&D leaders have begun to address this
issue by instituting different evaluation and
reward systems.
For example, Yamada observed that Sir
James Black, who had the key role in the
development of blockers and histamine
blockers, never received a major bonus
for his discoveries. As a consequence of
this awareness, Yamada established a new
reward system at GlaxoSmithKline: I set
up a program where once we had a proof
ofconcept molecule, wed form a team of
the top scientists in the company. It was
like a Nobel committee. Theyd go back
and research how was it we got here. If they
could identify one or two or three people
that were really the fundamental reason
for the success of that molecule, then those
people got an extraordinary deal. However,
although Armstrong admits that they should
find a way to reward the members of the
CHORUS that might be different from the
rest of R&D at Eli lilly, he remarked: In this
experiment, this group, first and foremost,
is hugely motivated, very energetic in large
part because theyre uniquely empowered to
actually carry on a tremendous amount of
work in the organization.
Theme 4: underutilized middle managers
I used to give a lecture, and still do, which is
entitled The power of the light you shine.
George Milne, former Executive Vice President
of Global Research and Development, Pfizer
The role of middle managers defined as
those individuals in large pharmaceutical
companies who are not at the executive level,
but who lead large groups or departments
and have considerable responsibilities for
the successful conduct of projects and
programmes emerged as an important
concern in the productivity of discovery
units, even more so than in the case of
development. Milne observed that scientists
become middle managers as a reward for
their outstanding scientific contributions
and often for their entrepreneurial behav
iour. However, after they attain the status
of middle managers, they are rewarded for
productivity as measured by the number
of compounds at the required phase, by
achieving timelines and by performing
activities that engender need for control
and predictability, none of which might
be the appropriate measures for discovery.
Consequently, middle managers can become
frustrated with this change in strategies.
Milne motivated middle managers through
his Hawthorneeffectstyle lecture, reminding
them that although they may not feel that
important, the scientists working under
them feel they are really important and, as
a result, could tell them where their light
shines and where it doesnt.
As we probed the desired qualities that
would facilitate the selection of middle
managers most likely to foster entrepre
neurial behaviour, Yamada, Ringrose,
Sato and Douglas identified two discovery
types, which we term mavericks and true
believers. Mavericks are those scientists
who, although they may share the goals of
the organization, seem to achieve superior
results by thinking and working outside
the prescribed or routine organizational
conventions. True believers not only are
convinced of the scientific approach of their
given project, but can also communicate and
recruit organizational support and resources
for the project a particularly important
quality, according to Sato.
Steven Paul, Armstrong, Boger and
Fishman stressed the importance of middle
managers retaining their scientific/technical
edge if the company is to benefit from
collaborations. Paul and Armstrong also
described a series of approaches used at
Eli lilly to engage middle managers in
entrepreneurial activities. They focused
on the sourcing of innovation as a means
of getting their middle and senior managers
to take ownership for finding the best
solutions, regardless of where they were
identified. Thus, lilly was an early adopter
of the practice of leveraging knowledge
from outside the company for solutions to
internal problems. This successful approach
was later spun out as Innocentive. lilly

Box 2 | Study methods
Before our interviews, we sent each interviewee our lists of behaviours that described both
individual and organizational entrepreneurship and the following questions to serve as a basis
for the interview:
general questions
Did entrepreneurial and big firm characteristics fuel the growth of large biopharmaceutical
companies (big pharma)?
Was or is there entrepreneurial behaviour in big pharma?
Were there lessons that can inform new organizational paradigms to increase productivity?
interview questions
How would you describe your R&D organization when you assumed leadership?
What were its strengths and weaknesses?
What were some of the organizational changes you introduced?
What did you hope to achieve?
How did you balance individual recognition versus team recognition?
What specific outcomes can you cite that were affected by the changes you introduced?
What are the differences between todays challenges and those that existed when you
were head of R&D?
What would you do today to address those challenges?
Did you create any special teams, spin-outs or collaborations to achieve your objectives?
How did you access R&D knowledge from outside your firm? How has this changed in
recent times?
How did you balance your portfolio with respect to investing in long-term projects versus
short-term projects?
The interviews were carried out by teleconference by F. L. Douglas and V. K. Narayanan and
were taped with the permission of the participants. This process allowed us to focus on the
interview and use the transcripts afterwards to improve our understanding of an issue or later
send questions for clarification to the interviewees.
PersPecti ves
also introduced the proofofconcept unit
CHORUS (see above), and in an extension of
this concept, lilly launched a joint venture
with Jubilant Organosys in Bangalore, India,
to develop molecules between preclinical
and Phase II testing in May 2009. Through
this collaboration, lilly also benefits from
the potential opportunity to have access to
successful proofofconcept compounds
from sources other than their own research
laboratories. Armstrong said that lilly has
expanded this concept to several companies
in India and China, who work on selected
compounds that lilly has the right to buy
back when predetermined milestones
are achieved. These activities dismantle
vestiges of the not invented here
syndrome that can mean certain death to
entrepreneurial behaviour, he noted.
Paul of Eli lilly summarizes this more
broadly: So, one element of the trans
formation from a corporate perspective
structurally is to go from this FIPCO (Fully
Integrated Pharmaceutical Company)
structure to a FIPNET, Fully Integrated
Pharmaceutical Network. Weve defined
three levels of this FIPNET. level one is
more traditional outsourcing. level two
leverages partnerships with universities and
small as well as larger companies. These
valuesharing partnerships provide both risk
sharing and cost sharing. Finally, a level
three FIPNET or partial ownership model
allows small companies to pursue projects
within the context of their unique cultures.
Our interviews also suggest that discovery
stage collaborations are perceived to be
effective when the collaborators share a
common sense of curiosity and focus on
science. The importance of curiosity as a
driver for the selection of technology, bio
logical mechanisms and collaborations was
stressed as a key component of the culture at
Genzyme by Smith, who added that curiosity
refers to a desire to learn and benefit from
knowledge that others in that university in
India or in France might deem interesting.
We determined that among the tasks
that need to be performed at the middle
management level are the maintenance of
organizational knowledge, the nurturing
of mavericks and true believers, and the
provision of support for the drug discovery
teams. Essential to these roles is the need
to remain close to science.
One of the distinguishing characteristics
of the biotechnology researchers we inter
viewed was their closeness to science and
their desire to retain this proximity. In large
pharmaceutical companies, however, exces
sive concern with milestones often detracts
from the ability of research teams to remain
close to science. Yet to the extent that the
rewards system associated with middle
management is based on the attainment
of milestones and not on the nurturing of
science, bureaucracy imposes barriers that
are detrimental to drug innovation. Indeed,
evolving a more sophisticated and nuanced
reward system for middle management may
be among the most important organizational
priorities for large pharmaceutical compa
nies to stimulate productivity in the crucial
discovery stage.
Theme 5: CEOHead of R&D interaction
You have to really be able to communicate
with your CEO if youre going to do your
job optimally in R&D. Gus Watanabe,
former Executive Vice President of Science
and Technology, Eli lilly
Several of the interviewees, including leon
Rosenberg, Watanabe, Goodman, Milne and
Glenn, stressed the importance of a close
alignment between the CEO and the head of
R&D in building and maintaining an optimal
culture in discovery. Boger described building
an innovative and productive organization as
more of a social experiment than a technical
one. He and Sato described in nearly identical
terms their tireless and consistent efforts to
build, nurture and maintain elements such
as interdisciplinary teamwork, rapid access
to information and a sense of ownership
in their organization. Rosenberg described
the weekly meetings that he used to have
with the CEO: He wasnt a scientist, but he
appreciated science, and he liked scientists.
He and I would converse regularly after
he would read his weekly copy of The New
England Journal of Medicine. This was a great
joy of mine because it kept me very close to
him. It allowed me to put science high in
the order of his thoughts. Smith described
similar efforts and alignment with Genzyme
CEO Henri Termeer in maintaining their
organizational culture. Busch and Bernd
Kirschbaum gave examples of the consider
able challenges and efforts needed to build a
common culture in a newly merged organi
zation and its impact on innovation, and
the important role that alignment with their
CEOs had in their efforts to build the desired
culture in discovery. Douglas also com
mented on the importance of the close align
ment between him and Richard Markham
when they took on the task of building
Hoechst Marion Roussel, the result of merg
ing Hoechst Pharmaceutical, Marion Merrell
Dow and Roussel Uclaf, in their respective
roles as Head of R&D and CEO.
Theme 6: technology and R&D models
My major responsibilities as given to me by
the CEO when I joined BMS were to forge a
single R&D organization out of two groups
with differing strengths in Discovery and
Development. leon Rosenberg, Professor at
Princeton University and former Senior Vice
President of Scientific Affairs, BristolMyers
Squibb
The interviews revealed the unexpected
finding that although each R&D leader
recognized that changing science and tech
nology influenced how they organized their
research units, most had not reflected on the
impact that organizational changes might
have on entrepreneurial behaviour. The
interviewees identified four types of organi
zational structures or models that defined
the point at which commitment was made
to the fullscale clinical development of a
compound (FIG. 1).
First, they described a traditional R&D
organization, in which the transition from
research to development occurred with
the filing of an investigational new drug
application. This is termed the preclinical
proofofconcept (pPoC) model. The second
structure mentioned the human proofof
concept (hPoC) model was similar to this,
with the addition of clinical pharmacology
to improve dose selection and the search for
efficacy in humans.
The third structure, known as the
clinical proofofconcept (cPoC) model,
includes Phase I and Phase IIa trials in
research. It uses biomarkers and pharmaco
kinetic studies to select the best compound
to determine efficacy in the target patients,
with efficacy in Phase IIa trials being the
crucial decision point for full development
(FIG. 1).
Watanabe, deceased since the interviews,
introduced this concept by having M.D.s or
M.D./Ph.D.s responsible for each discovery
therapeutic unit from the concept stage to
Phase IIa trials at Eli lilly. Glenn introduced
the clinical biology unit concept in research at
Ciba Geigy in 1984 with the goal of carrying
out specific studies in patients to achieve
proof of concept. Several other companies
have adopted this model in various forms.
These include the research and early devel
opment (RED) units at Johnson & Johnson,
the research units at MerckSerono, global
drug discovery at Bayer Schering Pharma
and disease biology areas at Roche (it
should be noted that Roches new focus on
personalized healthcare by interweaving
their diagnostic and pharmaceutical divi
sions has moved them to the fourth model;
PersPecti ves
Theme 3: reward systems matter
I am firmly a believer that you get what
you reward. Josh Boger, former CEO,
Vertex Pharmaceuticals
In most industries, as companies grow, there
is pressure for each to offer uniform com
pensation, rewards and benefits. Fishman
and Goodman, as well as Peter Ringrose,
each referred to the tendency for all depart
ments to compensate in a similar fashion as
a homogenizing effect. What is curious is
that although large companies recognize the
need to implement special bonus systems
to incentivize sale representatives to act
entrepreneurially, the opposite occurs in
R&D. The lack of incentives has persisted
owing to organizational inertia, but thought
ful R&D leaders have begun to address this
issue by instituting different evaluation and
reward systems.
For example, Yamada observed that Sir
James Black, who had the key role in the
development of blockers and histamine
blockers, never received a major bonus
for his discoveries. As a consequence of
this awareness, Yamada established a new
reward system at GlaxoSmithKline: I set
up a program where once we had a proof
ofconcept molecule, wed form a team of
the top scientists in the company. It was
like a Nobel committee. Theyd go back
and research how was it we got here. If they
could identify one or two or three people
that were really the fundamental reason
for the success of that molecule, then those
people got an extraordinary deal. However,
although Armstrong admits that they should
find a way to reward the members of the
CHORUS that might be different from the
rest of R&D at Eli lilly, he remarked: In this
experiment, this group, first and foremost,
is hugely motivated, very energetic in large
part because theyre uniquely empowered to
actually carry on a tremendous amount of
work in the organization.
Theme 4: underutilized middle managers
I used to give a lecture, and still do, which is
entitled The power of the light you shine.
George Milne, former Executive Vice President
of Global Research and Development, Pfizer
The role of middle managers defined as
those individuals in large pharmaceutical
companies who are not at the executive level,
but who lead large groups or departments
and have considerable responsibilities for
the successful conduct of projects and
programmes emerged as an important
concern in the productivity of discovery
units, even more so than in the case of
development. Milne observed that scientists
become middle managers as a reward for
their outstanding scientific contributions
and often for their entrepreneurial behav
iour. However, after they attain the status
of middle managers, they are rewarded for
productivity as measured by the number
of compounds at the required phase, by
achieving timelines and by performing
activities that engender need for control
and predictability, none of which might
be the appropriate measures for discovery.
Consequently, middle managers can become
frustrated with this change in strategies.
Milne motivated middle managers through
his Hawthorneeffectstyle lecture, reminding
them that although they may not feel that
important, the scientists working under
them feel they are really important and, as
a result, could tell them where their light
shines and where it doesnt.
As we probed the desired qualities that
would facilitate the selection of middle
managers most likely to foster entrepre
neurial behaviour, Yamada, Ringrose,
Sato and Douglas identified two discovery
types, which we term mavericks and true
believers. Mavericks are those scientists
who, although they may share the goals of
the organization, seem to achieve superior
results by thinking and working outside
the prescribed or routine organizational
conventions. True believers not only are
convinced of the scientific approach of their
given project, but can also communicate and
recruit organizational support and resources
for the project a particularly important
quality, according to Sato.
Steven Paul, Armstrong, Boger and
Fishman stressed the importance of middle
managers retaining their scientific/technical
edge if the company is to benefit from
collaborations. Paul and Armstrong also
described a series of approaches used at
Eli lilly to engage middle managers in
entrepreneurial activities. They focused
on the sourcing of innovation as a means
of getting their middle and senior managers
to take ownership for finding the best
solutions, regardless of where they were
identified. Thus, lilly was an early adopter
of the practice of leveraging knowledge
from outside the company for solutions to
internal problems. This successful approach
was later spun out as Innocentive. lilly

Box 2 | Study methods
Before our interviews, we sent each interviewee our lists of behaviours that described both
individual and organizational entrepreneurship and the following questions to serve as a basis
for the interview:
general questions
Did entrepreneurial and big firm characteristics fuel the growth of large biopharmaceutical
companies (big pharma)?
Was or is there entrepreneurial behaviour in big pharma?
Were there lessons that can inform new organizational paradigms to increase productivity?
interview questions
How would you describe your R&D organization when you assumed leadership?
What were its strengths and weaknesses?
What were some of the organizational changes you introduced?
What did you hope to achieve?
How did you balance individual recognition versus team recognition?
What specific outcomes can you cite that were affected by the changes you introduced?
What are the differences between todays challenges and those that existed when you
were head of R&D?
What would you do today to address those challenges?
Did you create any special teams, spin-outs or collaborations to achieve your objectives?
How did you access R&D knowledge from outside your firm? How has this changed in
recent times?
How did you balance your portfolio with respect to investing in long-term projects versus
short-term projects?
The interviews were carried out by teleconference by F. L. Douglas and V. K. Narayanan and
were taped with the permission of the participants. This process allowed us to focus on the
interview and use the transcripts afterwards to improve our understanding of an issue or later
send questions for clarification to the interviewees.
PersPecti ves
and integration of information generated
by all the relevant disciplines, regardless
of where it is found, must be assisted and
rewarded. We recommend that companies
seek to move towards the sPoC model
(FIG. 1), which extends research to Phase IIa
and also incorporates feedback from
postmarketing surveillance studies to
fuel the continued search for new targets,
bio markers and an understanding of
offtarget effects. These activities are
crucial in the search for the right therapy
for the right patient, and will heighten
the sense of urgency to get drugs to
patients.
Middle managers should be given
incentives to access the best ideas and
the best people globally to drive curiosity
and open innovation. Managing external
networks and rapidly internalizing the
results of new scientific and technical
approaches should also be among their
priorities. Rewards for managers should
be based on the speed with which they
achieve the progression of projects,
facilitated by their capacity to foster the
integration of knowledge and to provide
appropriate guidance to true believers
and mavericks. Ultimately, companies
should use middle managers to transform
their culture from something analogous
to a supertanker to a flotilla of diverse,
nimble and innovative ships, consisting
of the following elements: focus on the
patient, scientific curiosity, collaboration,
speed to solution and rewards that are
aligned with goals.
Multidisciplinary research teams should
consist of no more than 2040 members
who focus on the preclinical and clinical
validation of novel targets/mechanisms
with responsibility through to clinical
proofofconcept in Phase IIa trials.
These units of innovation should be
supported by the traditional discovery
research expert groups, such as molecular
biology and chemistry, which focus on
assessing the molecular validation or
relevance of the target and finding lead
molecules that interact with the target.
Both of these groups should include
specialists in postmarketing surveillance
to ensure relevance to the patient.
They should also perform relevant studies
with academic clinicians to generate
information from subsets of patients,
which might inform the research efforts
and ensure a sPoC approach.
Heads of R&D must focus indefatigably
on building and maintaining an
appropriate entrepreneurial culture.
Furthermore, the support of the CEO
for this focus must be visible and active.
Companies should be more innovative
with respect to reward and recognition
for discovery scientists, to foster bias for
action, ownership and an appropriate
sense of urgency.
Companies should examine what we
term the columns outside the doors
phenomenon and the subtle impact that
this form of recognition might have
on entrepreneurial behaviour. Smith
described this phenomenon, which
occurs across the world: as startup
companies become successful, they are
relocated from humble laboratories to
grander buildings with columns outside
their doors. Interestingly, such edifices
often violate the observed inverse
square relationship between com
munication among scientists in labo
ratories and the distance between these
laboratories. We offer this insight more
as a provocative thought than as a firm
recommendation.
We offer one final thought regarding the
transferability of these insights from drug
discovery research to drug development.
In discovery, our data and previous research
suggests that concept realization the
metric relevant for gauging the effectiveness
of discovery improves as the size of the
innovation unit grows to around 20 to 40
researchers, but then declines as the size of the
unit increases and diseconomies of scale take
effect. In development, it is common wisdom
that the optimal efficiency of the process is
obtained when the size of the unit is consider
ably larger than the optimal size of discovery
units. Interestingly, in large pharmaceutical
companies, it has become increasingly
common to outsource many development
related activities, and as this extends, it is
likely that the efficiencysize relationship
may be turned on its head, with pharma
ceutical companies managing development
with relatively smaller units than they have
at present. In that case, the management of
discovery units may hold valuable lessons
for the conduct of development. These
observations might stimulate large
pharmaceutical companies to do that which
Yamada maintains they can: transform
the industry.
Frank L. Douglas, Lesa Mitchell and Robert E. Litan
are at the Ewing Marion Kauffman Foundation,
Kansas City, Missouri, USA.
Frank L. Douglas is also at the Austen BioInnovation
Institute, Akron, Ohio, USA.
V. K. Narayanan is at the Center for Research
Excellence, LeBow College of Business,
Drexel University, Philadelphia, Pennsylvania, USA.
Correspondence to F.L.D.
e-mail: fdouglas@kauffman.org
doi:10.1038/nrd3230
Published online 20 August 2010
1. Goodman, M. Pharma industry performance metrics:
20072012E. Nature Rev. Drug Discov. 7, 795
(2008).
2. Pisano, G. P. Science Business: The Promise, the
Reality, and the Future of Biotech (Harvard Business
School Press, Boston, Massachusetts, USA, 2006).
3. Cuervo, A., Ribeiro, D. & Roig, S (eds)
Entrepreneurship: Concepts, Theory and Perspective
(Springer, Berlin, Germany, 2007).
4. Schramm, C. J. The Entrepreneurial Imperative:
How Americas Economic Miracle Will Reshape the
World (and Change Your Life) (HarperCollins, 2006).
5. Pharmaceutical Research and Manufacturers of
America. PhRMA Annual Membership Survey.
PhRMA website [online], http://www.phrma.org/files/
attachments/PhRMA%202009%20Profile%20
FINAL.pdf (2009).
6. IMS Health. Global Pharmaceutical Sales, 20012008.
IMS Health website [online], http://www.imshealth.
com/deployedfiles/imshealth/Global/Content/
StaticFile/Top_Line_Data/Global_Pharma_
Sales_2001-2008_Version_2.pdf (2009).
7. Munos, B. Lessons from 60 years of pharmaceutical
(2009).
Acknowledgements
With special thanks to D. Kaiser, M. Tribbitt and T. Flores for
their research and editorial assistance. We would also like to
thank the following colleagues who participated in these inter-
views: B. Adelman, R. Armstrong, L. Babiss, J. Boger, A. Busch,
P. Corr, M. Fishman, T. Glenn, C. Goodman, B. Kirschbaum,
J. Leiden, G. Milne, P. Needleman, G. Neil, J. Patterson,
S. Paul, J. Reinhardt, P. Ringrose, D. Rosen, L. Rosenberg,
R. Ruffolo, V. Sato, B. Shapiro, A. Smith, G. Watanabe and
T. Yamada (see Box 1).
FuRTHER INFORMATION
Austen Bioinnovation institution homepage:
www.abiakron.org
All links Are Active in the online pDf
PersPecti ves
Discovery
Discovery and
clinical pharmacology
Discovery, Phase I and Phase IIa trials
Discovery, Phase I and Phase IIa trials, and PMS
Preclinical PoC
Human PoC
Clinical PoC
Stratified PoC
Feedback
and patient
stratification
see below). The formation of the Novartis
Institutes for Biomedical Research (NIBR)
has served to institutionalize this practice,
and to implement the fourth model (see
below).
Present interest in the potential of
translational medicine to find the right
drug for the right patient requires the
integration of several activities. These include
the search for biomarkers that not only help
to select compounds preclinically, but also
help to predict, stratify and monitor the
patients and subgroups who will experience
the requisite efficacy of the compound
and an acceptable level of adverse events.
To achieve this, collaboration between
research and postmarketing surveillance
groups is necessary. The increasing focus
on adverse events, which often have a low
frequency in the preregistration trials
(Phase IIb and Phase III), argues for the
fourth organizational model: the stratified
proofofconcept (sPoC) model (FIG. 1).
This structure allows active feedback
between discovery research, Phase I and
Phase IIa trials and postmarketing
surveillance, and includes the potential
development of biomarkers to identify
those patients who might be more likely
to experience adverse events. There are
instances with such models in which a
change in the administration regimen of
the drug or the selection of an alternative
drug has provided greater benefit to
particular patients.
The impact of structure on the type of
organization is probably best understood
by some historical examples. In 1987, Pfizer
was an early adopter of the clinical biology
unit concept, in which the focus was on
identifying markers of disease and human
models of disease that would rapidly deter
mine whether, and in which patients, novel
mechanismbased compounds would be
efficacious. However, the traditional R&D
structure at Pfizer, in which the transition
between research and development is made
at the beginning of human testing (Phase I),
probably delayed the full implementation of a
cPoC type organization. By contrast, organi
zations such as Eli lilly, which for many
years included Phase I and Phase IIa in the
research units, easily functioned as cPoC type
organizations and are making the transition
to sPoC organizations more rapidly. Another
example is that of Novartis, in which the
creation of the NIBR rapidly converted this
organization from a cPoC to a sPoC focus.
Finally, at Genzyme, the unique focus and
the involvement of patients in the pursuit of
drug innovation contributed to the company
becoming one of the earliest adopters of a
sPoC focus.
Of particular interest is the realization
that the heads of R&D who embrace the
sPoC model all seek to generate a bias for
action that places the patient at the centre
of the research effort. Navigation through
the complexity of priorities in an increas
ingly large company is often the principal
challenge.
Recommendations
Establishing and maintaining an entrepre
neurial culture during drug development
is perhaps easier than during the research
stage as the activities associated with this
later phase are closer to product realization
and clinical application. The inherent sense
of urgency associated with the development
stage is driven by the competitive environ
ment, in which time to market with a
differentiated product is an important
determinant of success. Thus, the creation of
special product teams for individual efforts
and for therapeutic franchises, which are
coled by development and commercial
leaders, is more likely to foster the entrepre
neurial behaviours we have identified, such as
ownership, outcome focus, passion and con
viction, and the ability to recruit the best
people. During this stage, it is also easier to
define successful outcomes, such as comple
tion of a welldefined task in a rapid time
frame, and so it is easier to create incentives
that directly reward achieving these outcomes.
During the research stage, however,
entrepreneurial behaviours are often
compromised by several characteristics.
These include:
Increasing size and complexity of
research groups
Having large portfolios and a focus on
increasing the number of shots on goal
Middle managers focused on timelines
and portfolios, instead of science, technol
ogy and leveraging external knowledge
Influence of the commercial department
too early in the process
The impact of evolving science and
technology on organizational complexity
A lack of alignment between Head of
R&D and CEO with respect to the culture
of research.
The responses from our interviewees with
respect to these differences led us to the
following recommendations:
Organizational structures in research units
should facilitate identifying the right drug
for the right patient. The rapid sharing
Figure 1 | evolution of discovery models to identify the right drug for the right patient. Four
types of organizational structures for discovery research were identified on the basis of the point at
which a commitment is made to the full-scale clinical development of a compound. the integration
of activities needed to identify the right drug for the right patient is catalysing a transition towards
the stratified proof-of-concept (Poc) model, which allows active feedback between discovery,
research, Phase i and Phase iia trials, and post-marketing surveillance (PMs).
PersPecti ves
and integration of information generated
by all the relevant disciplines, regardless
of where it is found, must be assisted and
rewarded. We recommend that companies
seek to move towards the sPoC model
(FIG. 1), which extends research to Phase IIa
and also incorporates feedback from
postmarketing surveillance studies to
fuel the continued search for new targets,
bio markers and an understanding of
offtarget effects. These activities are
crucial in the search for the right therapy
for the right patient, and will heighten
the sense of urgency to get drugs to
patients.
Middle managers should be given
incentives to access the best ideas and
the best people globally to drive curiosity
and open innovation. Managing external
networks and rapidly internalizing the
results of new scientific and technical
approaches should also be among their
priorities. Rewards for managers should
be based on the speed with which they
achieve the progression of projects,
facilitated by their capacity to foster the
integration of knowledge and to provide
appropriate guidance to true believers
and mavericks. Ultimately, companies
should use middle managers to transform
their culture from something analogous
to a supertanker to a flotilla of diverse,
nimble and innovative ships, consisting
of the following elements: focus on the
patient, scientific curiosity, collaboration,
speed to solution and rewards that are
aligned with goals.
Multidisciplinary research teams should
consist of no more than 2040 members
who focus on the preclinical and clinical
validation of novel targets/mechanisms
with responsibility through to clinical
proofofconcept in Phase IIa trials.
These units of innovation should be
supported by the traditional discovery
research expert groups, such as molecular
biology and chemistry, which focus on
assessing the molecular validation or
relevance of the target and finding lead
molecules that interact with the target.
Both of these groups should include
specialists in postmarketing surveillance
to ensure relevance to the patient.
They should also perform relevant studies
with academic clinicians to generate
information from subsets of patients,
which might inform the research efforts
and ensure a sPoC approach.
Heads of R&D must focus indefatigably
on building and maintaining an
appropriate entrepreneurial culture.
Furthermore, the support of the CEO
for this focus must be visible and active.
Companies should be more innovative
with respect to reward and recognition
for discovery scientists, to foster bias for
action, ownership and an appropriate
sense of urgency.
Companies should examine what we
term the columns outside the doors
phenomenon and the subtle impact that
this form of recognition might have
on entrepreneurial behaviour. Smith
described this phenomenon, which
occurs across the world: as startup
companies become successful, they are
relocated from humble laboratories to
grander buildings with columns outside
their doors. Interestingly, such edifices
often violate the observed inverse
square relationship between com
munication among scientists in labo
ratories and the distance between these
laboratories. We offer this insight more
as a provocative thought than as a firm
recommendation.
We offer one final thought regarding the
transferability of these insights from drug
discovery research to drug development.
In discovery, our data and previous research
suggests that concept realization the
metric relevant for gauging the effectiveness
of discovery improves as the size of the
innovation unit grows to around 20 to 40
researchers, but then declines as the size of the
unit increases and diseconomies of scale take
effect. In development, it is common wisdom
that the optimal efficiency of the process is
obtained when the size of the unit is consider
ably larger than the optimal size of discovery
units. Interestingly, in large pharmaceutical
companies, it has become increasingly
common to outsource many development
related activities, and as this extends, it is
likely that the efficiencysize relationship
may be turned on its head, with pharma
ceutical companies managing development
with relatively smaller units than they have
at present. In that case, the management of
discovery units may hold valuable lessons
for the conduct of development. These
observations might stimulate large
pharmaceutical companies to do that which
Yamada maintains they can: transform
the industry.
Frank L. Douglas, Lesa Mitchell and Robert E. Litan
are at the Ewing Marion Kauffman Foundation,
Kansas City, Missouri, USA.
Frank L. Douglas is also at the Austen BioInnovation
Institute, Akron, Ohio, USA.
V. K. Narayanan is at the Center for Research
Excellence, LeBow College of Business,
Drexel University, Philadelphia, Pennsylvania, USA.
Correspondence to F.L.D.
e-mail: fdouglas@kauffman.org
doi:10.1038/nrd3230
Published online 20 August 2010
1. Goodman, M. Pharma industry performance metrics:
20072012E. Nature Rev. Drug Discov. 7, 795
(2008).
2. Pisano, G. P. Science Business: The Promise, the
Reality, and the Future of Biotech (Harvard Business
School Press, Boston, Massachusetts, USA, 2006).
3. Cuervo, A., Ribeiro, D. & Roig, S (eds)
Entrepreneurship: Concepts, Theory and Perspective
(Springer, Berlin, Germany, 2007).
4. Schramm, C. J. The Entrepreneurial Imperative:
How Americas Economic Miracle Will Reshape the
World (and Change Your Life) (HarperCollins, 2006).
5. Pharmaceutical Research and Manufacturers of
America. PhRMA Annual Membership Survey.
PhRMA website [online], http://www.phrma.org/files/
attachments/PhRMA%202009%20Profile%20
FINAL.pdf (2009).
6. IMS Health. Global Pharmaceutical Sales, 20012008.
IMS Health website [online], http://www.imshealth.
com/deployedfiles/imshealth/Global/Content/
StaticFile/Top_Line_Data/Global_Pharma_
Sales_2001-2008_Version_2.pdf (2009).
7. Munos, B. Lessons from 60 years of pharmaceutical
(2009).
Acknowledgements
With special thanks to D. Kaiser, M. Tribbitt and T. Flores for
their research and editorial assistance. We would also like to
thank the following colleagues who participated in these inter-
views: B. Adelman, R. Armstrong, L. Babiss, J. Boger, A. Busch,
P. Corr, M. Fishman, T. Glenn, C. Goodman, B. Kirschbaum,
J. Leiden, G. Milne, P. Needleman, G. Neil, J. Patterson,
S. Paul, J. Reinhardt, P. Ringrose, D. Rosen, L. Rosenberg,
R. Ruffolo, V. Sato, B. Shapiro, A. Smith, G. Watanabe and
T. Yamada (see Box 1).
FuRTHER INFORMATION
Austen Bioinnovation institution homepage:
www.abiakron.org
All links Are Active in the online pDf
PersPecti ves
Discovery
Discovery and
clinical pharmacology
Discovery, Phase I and Phase IIa trials
Discovery, Phase I and Phase IIa trials, and PMS
Preclinical PoC
Human PoC
Clinical PoC
Stratified PoC
Feedback
and patient
stratification
see below). The formation of the Novartis
Institutes for Biomedical Research (NIBR)
has served to institutionalize this practice,
and to implement the fourth model (see
below).
Present interest in the potential of
translational medicine to find the right
drug for the right patient requires the
integration of several activities. These include
the search for biomarkers that not only help
to select compounds preclinically, but also
help to predict, stratify and monitor the
patients and subgroups who will experience
the requisite efficacy of the compound
and an acceptable level of adverse events.
To achieve this, collaboration between
research and postmarketing surveillance
groups is necessary. The increasing focus
on adverse events, which often have a low
frequency in the preregistration trials
(Phase IIb and Phase III), argues for the
fourth organizational model: the stratified
proofofconcept (sPoC) model (FIG. 1).
This structure allows active feedback
between discovery research, Phase I and
Phase IIa trials and postmarketing
surveillance, and includes the potential
development of biomarkers to identify
those patients who might be more likely
to experience adverse events. There are
instances with such models in which a
change in the administration regimen of
the drug or the selection of an alternative
drug has provided greater benefit to
particular patients.
The impact of structure on the type of
organization is probably best understood
by some historical examples. In 1987, Pfizer
was an early adopter of the clinical biology
unit concept, in which the focus was on
identifying markers of disease and human
models of disease that would rapidly deter
mine whether, and in which patients, novel
mechanismbased compounds would be
efficacious. However, the traditional R&D
structure at Pfizer, in which the transition
between research and development is made
at the beginning of human testing (Phase I),
probably delayed the full implementation of a
cPoC type organization. By contrast, organi
zations such as Eli lilly, which for many
years included Phase I and Phase IIa in the
research units, easily functioned as cPoC type
organizations and are making the transition
to sPoC organizations more rapidly. Another
example is that of Novartis, in which the
creation of the NIBR rapidly converted this
organization from a cPoC to a sPoC focus.
Finally, at Genzyme, the unique focus and
the involvement of patients in the pursuit of
drug innovation contributed to the company
becoming one of the earliest adopters of a
sPoC focus.
Of particular interest is the realization
that the heads of R&D who embrace the
sPoC model all seek to generate a bias for
action that places the patient at the centre
of the research effort. Navigation through
the complexity of priorities in an increas
ingly large company is often the principal
challenge.
Recommendations
Establishing and maintaining an entrepre
neurial culture during drug development
is perhaps easier than during the research
stage as the activities associated with this
later phase are closer to product realization
and clinical application. The inherent sense
of urgency associated with the development
stage is driven by the competitive environ
ment, in which time to market with a
differentiated product is an important
determinant of success. Thus, the creation of
special product teams for individual efforts
and for therapeutic franchises, which are
coled by development and commercial
leaders, is more likely to foster the entrepre
neurial behaviours we have identified, such as
ownership, outcome focus, passion and con
viction, and the ability to recruit the best
people. During this stage, it is also easier to
define successful outcomes, such as comple
tion of a welldefined task in a rapid time
frame, and so it is easier to create incentives
that directly reward achieving these outcomes.
During the research stage, however,
entrepreneurial behaviours are often
compromised by several characteristics.
These include:
Increasing size and complexity of
research groups
Having large portfolios and a focus on
increasing the number of shots on goal
Middle managers focused on timelines
and portfolios, instead of science, technol
ogy and leveraging external knowledge
Influence of the commercial department
too early in the process
The impact of evolving science and
technology on organizational complexity
A lack of alignment between Head of
R&D and CEO with respect to the culture
of research.
The responses from our interviewees with
respect to these differences led us to the
following recommendations:
Organizational structures in research units
should facilitate identifying the right drug
for the right patient. The rapid sharing
Figure 1 | evolution of discovery models to identify the right drug for the right patient. Four
types of organizational structures for discovery research were identified on the basis of the point at
which a commitment is made to the full-scale clinical development of a compound. the integration
of activities needed to identify the right drug for the right patient is catalysing a transition towards
the stratified proof-of-concept (Poc) model, which allows active feedback between discovery,
research, Phase i and Phase iia trials, and post-marketing surveillance (PMs).
PersPecti ves
Were Up for the
Innovation Challenge.

We at Eli Lilly and Company are on
a mission: to optimize cutting edge
science and technology in order to
develop medicines that help people live
longer, healthier and more active lives.
Its a legacy that began for us 135 years
ago in a small building in Indianapolis
and has expanded globally. Today,
Lilly delivers medicines to millions of
patients in 143 countries.
Discovering and developing a new
medicine has always been a highly
complex endeavor, often requiring
clever, breakthrough solutions. Today
the challenge is even greater. We need
to draw on all the brainpower and
creativity we can, no matter where in
the world that expertise resides.
That is one of the reasons we are
strong advocates of collaboration,
both within Lilly and with our external
partners.
Throughout Lillys history, advances
in human health have been achieved
because an individual scientist, or
a team of scientists, looked beyond
what was easy, familiar and safe.
True innovation requires us to solve
daunting scientifc problems and,
often, explore unknown territory. It
means that each of us must be willing
to stretch and challenge ourselves
every day. Lilly scientists dont avoid
diffcult scientifc problems because
they involve greater risk. We take
the risk openly and as a team. In this
way, we can continue to fnd unique
medicines that deliver improved
outcomes for individual patients.

Research is the Heart of the business,
the Soul of the enterprise.
Mr. Eli Lilly, grandson of the company founder,
Colonel Eli Lilly
clinical collections
reprint collection
collections
www.nature.com/reprintcollections/lilly/drug-innovation

00 The Future of Drug Innovation

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

00 The Future of Drug Innovation

Uploaded by

Copyright:

Available Formats

The future of drug innovation

Supplement to Nature Publishing Group Journals October 2011

iRND3 (Institute for Rare and

and Jason Anthony*

iRND3 (Institute for Rare and

and Jason Anthony*

These medicines have been withdrawn from the market.

These medicines have been withdrawn from the market.

Fulvestrant acts by promoting receptor degradation.

Fulvestrant acts by promoting receptor degradation.

channels are activated by RU-0211.

channels are activated by RU-0211.

in 1984, but until it was amended in 1992,

in 1984, but until it was amended in 1992,

You might also like